phoenix-oss/llama-stack-mirror
0
0
Fork 1
mirror of https://github.com/meta-llama/llama-stack.git synced 2025-10-10 21:34:36 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

feat: reuse previous mcp tool listings where possible (#3710)

Browse source 
# What does this PR do?
This PR checks whether, if a previous response is linked, there are
mcp_list_tools objects that can be reused instead of listing the tools
explicitly every time.

 Closes #3106 

## Test Plan
Tested manually.
Added unit tests to cover new behaviour.

---------

Signed-off-by: Gordon Sim <gsim@redhat.com>
Co-authored-by: Ashwin Bharambe <ashwin.bharambe@gmail.com>
This commit is contained in:
grs 2025-10-10 17:28:25 +01:00 • committed by GitHub
parent 0066d986c5
commit 8bf07f91cb
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
12 changed files with 1835 additions and 983 deletions
Download patch file Download diff file Expand all files Collapse all files

411
docs/static/deprecated-llama-stack-spec.html vendored
View file

@ -8975,6 +8975,168 @@
"title": "OpenAIResponseInputMessageContentText",
"description": "Text content for input messages in OpenAI response format."
},
"OpenAIResponseInputToolFileSearch": {
"type": "object",
"properties": {
"type": {
"type": "string",
"const": "file_search",
"default": "file_search",
"description": "Tool type identifier, always \"file_search\""
},
"vector_store_ids": {
"type": "array",
"items": {
"type": "string"
},
"description": "List of vector store identifiers to search within"
},
"filters": {
"type": "object",
"additionalProperties": {
"oneOf": [
{
"type": "null"
},
{
"type": "boolean"
},
{
"type": "number"
},
{
"type": "string"
},
{
"type": "array"
},
{
"type": "object"
}
]
},
"description": "(Optional) Additional filters to apply to the search"
},
"max_num_results": {
"type": "integer",
"default": 10,
"description": "(Optional) Maximum number of search results to return (1-50)"
},
"ranking_options": {
"type": "object",
"properties": {
"ranker": {
"type": "string",
"description": "(Optional) Name of the ranking algorithm to use"
},
"score_threshold": {
"type": "number",
"default": 0.0,
"description": "(Optional) Minimum relevance score threshold for results"
}
},
"additionalProperties": false,
"description": "(Optional) Options for ranking and scoring search results"
}
},
"additionalProperties": false,
"required": [
"type",
"vector_store_ids"
],
"title": "OpenAIResponseInputToolFileSearch",
"description": "File search tool configuration for OpenAI response inputs."
},
"OpenAIResponseInputToolFunction": {
"type": "object",
"properties": {
"type": {
"type": "string",
"const": "function",
"default": "function",
"description": "Tool type identifier, always \"function\""
},
"name": {
"type": "string",
"description": "Name of the function that can be called"
},
"description": {
"type": "string",
"description": "(Optional) Description of what the function does"
},
"parameters": {
"type": "object",
"additionalProperties": {
"oneOf": [
{
"type": "null"
},
{
"type": "boolean"
},
{
"type": "number"
},
{
"type": "string"
},
{
"type": "array"
},
{
"type": "object"
}
]
},
"description": "(Optional) JSON schema defining the function's parameters"
},
"strict": {
"type": "boolean",
"description": "(Optional) Whether to enforce strict parameter validation"
}
},
"additionalProperties": false,
"required": [
"type",
"name"
],
"title": "OpenAIResponseInputToolFunction",
"description": "Function tool configuration for OpenAI response inputs."
},
"OpenAIResponseInputToolWebSearch": {
"type": "object",
"properties": {
"type": {
"oneOf": [
{
"type": "string",
"const": "web_search"
},
{
"type": "string",
"const": "web_search_preview"
},
{
"type": "string",
"const": "web_search_preview_2025_03_11"
}
],
"default": "web_search",
"description": "Web search tool type variant to use"
},
"search_context_size": {
"type": "string",
"default": "medium",
"description": "(Optional) Size of search context, must be \"low\", \"medium\", or \"high\""
}
},
"additionalProperties": false,
"required": [
"type"
],
"title": "OpenAIResponseInputToolWebSearch",
"description": "Web search tool configuration for OpenAI response inputs."
},
"OpenAIResponseMCPApprovalRequest": {
"type": "object",
"properties": {
@ -9157,6 +9319,13 @@
"type": "number",
"description": "(Optional) Nucleus sampling parameter used for generation"
},
"tools": {
"type": "array",
"items": {
"$ref": "#/components/schemas/OpenAIResponseTool"
},
"description": "(Optional) An array of tools the model may call while generating a response."
},
"truncation": {
"type": "string",
"description": "(Optional) Truncation strategy applied to the response"
@ -9610,6 +9779,79 @@
"title": "OpenAIResponseText",
"description": "Text response configuration for OpenAI responses."
},
"OpenAIResponseTool": {
"oneOf": [
{
"$ref": "#/components/schemas/OpenAIResponseInputToolWebSearch"
},
{
"$ref": "#/components/schemas/OpenAIResponseInputToolFileSearch"
},
{
"$ref": "#/components/schemas/OpenAIResponseInputToolFunction"
},
{
"$ref": "#/components/schemas/OpenAIResponseToolMCP"
}
],
"discriminator": {
"propertyName": "type",
"mapping": {
"web_search": "#/components/schemas/OpenAIResponseInputToolWebSearch",
"file_search": "#/components/schemas/OpenAIResponseInputToolFileSearch",
"function": "#/components/schemas/OpenAIResponseInputToolFunction",
"mcp": "#/components/schemas/OpenAIResponseToolMCP"
}
}
},
"OpenAIResponseToolMCP": {
"type": "object",
"properties": {
"type": {
"type": "string",
"const": "mcp",
"default": "mcp",
"description": "Tool type identifier, always \"mcp\""
},
"server_label": {
"type": "string",
"description": "Label to identify this MCP server"
},
"allowed_tools": {
"oneOf": [
{
"type": "array",
"items": {
"type": "string"
}
},
{
"type": "object",
"properties": {
"tool_names": {
"type": "array",
"items": {
"type": "string"
},
"description": "(Optional) List of specific tool names that are allowed"
}
},
"additionalProperties": false,
"title": "AllowedToolsFilter",
"description": "Filter configuration for restricting which MCP tools can be used."
}
],
"description": "(Optional) Restriction on which tools can be used from this server"
}
},
"additionalProperties": false,
"required": [
"type",
"server_label"
],
"title": "OpenAIResponseToolMCP",
"description": "Model Context Protocol (MCP) tool configuration for OpenAI response object."
},
"OpenAIResponseUsage": {
"type": "object",
"properties": {
@ -9697,134 +9939,6 @@
}
}
},
"OpenAIResponseInputToolFileSearch": {
"type": "object",
"properties": {
"type": {
"type": "string",
"const": "file_search",
"default": "file_search",
"description": "Tool type identifier, always \"file_search\""
},
"vector_store_ids": {
"type": "array",
"items": {
"type": "string"
},
"description": "List of vector store identifiers to search within"
},
"filters": {
"type": "object",
"additionalProperties": {
"oneOf": [
{
"type": "null"
},
{
"type": "boolean"
},
{
"type": "number"
},
{
"type": "string"
},
{
"type": "array"
},
{
"type": "object"
}
]
},
"description": "(Optional) Additional filters to apply to the search"
},
"max_num_results": {
"type": "integer",
"default": 10,
"description": "(Optional) Maximum number of search results to return (1-50)"
},
"ranking_options": {
"type": "object",
"properties": {
"ranker": {
"type": "string",
"description": "(Optional) Name of the ranking algorithm to use"
},
"score_threshold": {
"type": "number",
"default": 0.0,
"description": "(Optional) Minimum relevance score threshold for results"
}
},
"additionalProperties": false,
"description": "(Optional) Options for ranking and scoring search results"
}
},
"additionalProperties": false,
"required": [
"type",
"vector_store_ids"
],
"title": "OpenAIResponseInputToolFileSearch",
"description": "File search tool configuration for OpenAI response inputs."
},
"OpenAIResponseInputToolFunction": {
"type": "object",
"properties": {
"type": {
"type": "string",
"const": "function",
"default": "function",
"description": "Tool type identifier, always \"function\""
},
"name": {
"type": "string",
"description": "Name of the function that can be called"
},
"description": {
"type": "string",
"description": "(Optional) Description of what the function does"
},
"parameters": {
"type": "object",
"additionalProperties": {
"oneOf": [
{
"type": "null"
},
{
"type": "boolean"
},
{
"type": "number"
},
{
"type": "string"
},
{
"type": "array"
},
{
"type": "object"
}
]
},
"description": "(Optional) JSON schema defining the function's parameters"
},
"strict": {
"type": "boolean",
"description": "(Optional) Whether to enforce strict parameter validation"
}
},
"additionalProperties": false,
"required": [
"type",
"name"
],
"title": "OpenAIResponseInputToolFunction",
"description": "Function tool configuration for OpenAI response inputs."
},
"OpenAIResponseInputToolMCP": {
"type": "object",
"properties": {
@ -9941,40 +10055,6 @@
"title": "OpenAIResponseInputToolMCP",
"description": "Model Context Protocol (MCP) tool configuration for OpenAI response inputs."
},
"OpenAIResponseInputToolWebSearch": {
"type": "object",
"properties": {
"type": {
"oneOf": [
{
"type": "string",
"const": "web_search"
},
{
"type": "string",
"const": "web_search_preview"
},
{
"type": "string",
"const": "web_search_preview_2025_03_11"
}
],
"default": "web_search",
"description": "Web search tool type variant to use"
},
"search_context_size": {
"type": "string",
"default": "medium",
"description": "(Optional) Size of search context, must be \"low\", \"medium\", or \"high\""
}
},
"additionalProperties": false,
"required": [
"type"
],
"title": "OpenAIResponseInputToolWebSearch",
"description": "Web search tool configuration for OpenAI response inputs."
},
"CreateOpenaiResponseRequest": {
"type": "object",
"properties": {
@ -10096,6 +10176,13 @@
"type": "number",
"description": "(Optional) Nucleus sampling parameter used for generation"
},
"tools": {
"type": "array",
"items": {
"$ref": "#/components/schemas/OpenAIResponseTool"
},
"description": "(Optional) An array of tools the model may call while generating a response."
},
"truncation": {
"type": "string",
"description": "(Optional) Truncation strategy applied to the response"

294
docs/static/deprecated-llama-stack-spec.yaml vendored
View file

@ -6661,6 +6661,122 @@ components:
title: OpenAIResponseInputMessageContentText
description: >-
Text content for input messages in OpenAI response format.
OpenAIResponseInputToolFileSearch:
type: object
properties:
type:
type: string
const: file_search
default: file_search
description: >-
Tool type identifier, always "file_search"
vector_store_ids:
type: array
items:
type: string
description: >-
List of vector store identifiers to search within
filters:
type: object
additionalProperties:
oneOf:
- type: 'null'
- type: boolean
- type: number
- type: string
- type: array
- type: object
description: >-
(Optional) Additional filters to apply to the search
max_num_results:
type: integer
default: 10
description: >-
(Optional) Maximum number of search results to return (1-50)
ranking_options:
type: object
properties:
ranker:
type: string
description: >-
(Optional) Name of the ranking algorithm to use
score_threshold:
type: number
default: 0.0
description: >-
(Optional) Minimum relevance score threshold for results
additionalProperties: false
description: >-
(Optional) Options for ranking and scoring search results
additionalProperties: false
required:
- type
- vector_store_ids
title: OpenAIResponseInputToolFileSearch
description: >-
File search tool configuration for OpenAI response inputs.
OpenAIResponseInputToolFunction:
type: object
properties:
type:
type: string
const: function
default: function
description: Tool type identifier, always "function"
name:
type: string
description: Name of the function that can be called
description:
type: string
description: >-
(Optional) Description of what the function does
parameters:
type: object
additionalProperties:
oneOf:
- type: 'null'
- type: boolean
- type: number
- type: string
- type: array
- type: object
description: >-
(Optional) JSON schema defining the function's parameters
strict:
type: boolean
description: >-
(Optional) Whether to enforce strict parameter validation
additionalProperties: false
required:
- type
- name
title: OpenAIResponseInputToolFunction
description: >-
Function tool configuration for OpenAI response inputs.
OpenAIResponseInputToolWebSearch:
type: object
properties:
type:
oneOf:
- type: string
const: web_search
- type: string
const: web_search_preview
- type: string
const: web_search_preview_2025_03_11
default: web_search
description: Web search tool type variant to use
search_context_size:
type: string
default: medium
description: >-
(Optional) Size of search context, must be "low", "medium", or "high"
additionalProperties: false
required:
- type
title: OpenAIResponseInputToolWebSearch
description: >-
Web search tool configuration for OpenAI response inputs.
OpenAIResponseMCPApprovalRequest:
type: object
properties:
@ -6802,6 +6918,12 @@ components:
type: number
description: >-
(Optional) Nucleus sampling parameter used for generation
tools:
type: array
items:
$ref: '#/components/schemas/OpenAIResponseTool'
description: >-
(Optional) An array of tools the model may call while generating a response.
truncation:
type: string
description: >-
@ -7158,6 +7280,56 @@ components:
title: OpenAIResponseText
description: >-
Text response configuration for OpenAI responses.
OpenAIResponseTool:
oneOf:
- $ref: '#/components/schemas/OpenAIResponseInputToolWebSearch'
- $ref: '#/components/schemas/OpenAIResponseInputToolFileSearch'
- $ref: '#/components/schemas/OpenAIResponseInputToolFunction'
- $ref: '#/components/schemas/OpenAIResponseToolMCP'
discriminator:
propertyName: type
mapping:
web_search: '#/components/schemas/OpenAIResponseInputToolWebSearch'
file_search: '#/components/schemas/OpenAIResponseInputToolFileSearch'
function: '#/components/schemas/OpenAIResponseInputToolFunction'
mcp: '#/components/schemas/OpenAIResponseToolMCP'
OpenAIResponseToolMCP:
type: object
properties:
type:
type: string
const: mcp
default: mcp
description: Tool type identifier, always "mcp"
server_label:
type: string
description: Label to identify this MCP server
allowed_tools:
oneOf:
- type: array
items:
type: string
- type: object
properties:
tool_names:
type: array
items:
type: string
description: >-
(Optional) List of specific tool names that are allowed
additionalProperties: false
title: AllowedToolsFilter
description: >-
Filter configuration for restricting which MCP tools can be used.
description: >-
(Optional) Restriction on which tools can be used from this server
additionalProperties: false
required:
- type
- server_label
title: OpenAIResponseToolMCP
description: >-
Model Context Protocol (MCP) tool configuration for OpenAI response object.
OpenAIResponseUsage:
type: object
properties:
@ -7219,98 +7391,6 @@ components:
file_search: '#/components/schemas/OpenAIResponseInputToolFileSearch'
function: '#/components/schemas/OpenAIResponseInputToolFunction'
mcp: '#/components/schemas/OpenAIResponseInputToolMCP'
OpenAIResponseInputToolFileSearch:
type: object
properties:
type:
type: string
const: file_search
default: file_search
description: >-
Tool type identifier, always "file_search"
vector_store_ids:
type: array
items:
type: string
description: >-
List of vector store identifiers to search within
filters:
type: object
additionalProperties:
oneOf:
- type: 'null'
- type: boolean
- type: number
- type: string
- type: array
- type: object
description: >-
(Optional) Additional filters to apply to the search
max_num_results:
type: integer
default: 10
description: >-
(Optional) Maximum number of search results to return (1-50)
ranking_options:
type: object
properties:
ranker:
type: string
description: >-
(Optional) Name of the ranking algorithm to use
score_threshold:
type: number
default: 0.0
description: >-
(Optional) Minimum relevance score threshold for results
additionalProperties: false
description: >-
(Optional) Options for ranking and scoring search results
additionalProperties: false
required:
- type
- vector_store_ids
title: OpenAIResponseInputToolFileSearch
description: >-
File search tool configuration for OpenAI response inputs.
OpenAIResponseInputToolFunction:
type: object
properties:
type:
type: string
const: function
default: function
description: Tool type identifier, always "function"
name:
type: string
description: Name of the function that can be called
description:
type: string
description: >-
(Optional) Description of what the function does
parameters:
type: object
additionalProperties:
oneOf:
- type: 'null'
- type: boolean
- type: number
- type: string
- type: array
- type: object
description: >-
(Optional) JSON schema defining the function's parameters
strict:
type: boolean
description: >-
(Optional) Whether to enforce strict parameter validation
additionalProperties: false
required:
- type
- name
title: OpenAIResponseInputToolFunction
description: >-
Function tool configuration for OpenAI response inputs.
OpenAIResponseInputToolMCP:
type: object
properties:
@ -7392,30 +7472,6 @@ components:
title: OpenAIResponseInputToolMCP
description: >-
Model Context Protocol (MCP) tool configuration for OpenAI response inputs.
OpenAIResponseInputToolWebSearch:
type: object
properties:
type:
oneOf:
- type: string
const: web_search
- type: string
const: web_search_preview
- type: string
const: web_search_preview_2025_03_11
default: web_search
description: Web search tool type variant to use
search_context_size:
type: string
default: medium
description: >-
(Optional) Size of search context, must be "low", "medium", or "high"
additionalProperties: false
required:
- type
title: OpenAIResponseInputToolWebSearch
description: >-
Web search tool configuration for OpenAI response inputs.
CreateOpenaiResponseRequest:
type: object
properties:
@ -7516,6 +7572,12 @@ components:
type: number
description: >-
(Optional) Nucleus sampling parameter used for generation
tools:
type: array
items:
$ref: '#/components/schemas/OpenAIResponseTool'
description: >-
(Optional) An array of tools the model may call while generating a response.
truncation:
type: string
description: >-

411
docs/static/llama-stack-spec.html vendored
View file

@ -7445,6 +7445,168 @@
"title": "OpenAIResponseInputFunctionToolCallOutput",
"description": "This represents the output of a function call that gets passed back to the model."
},
"OpenAIResponseInputToolFileSearch": {
"type": "object",
"properties": {
"type": {
"type": "string",
"const": "file_search",
"default": "file_search",
"description": "Tool type identifier, always \"file_search\""
},
"vector_store_ids": {
"type": "array",
"items": {
"type": "string"
},
"description": "List of vector store identifiers to search within"
},
"filters": {
"type": "object",
"additionalProperties": {
"oneOf": [
{
"type": "null"
},
{
"type": "boolean"
},
{
"type": "number"
},
{
"type": "string"
},
{
"type": "array"
},
{
"type": "object"
}
]
},
"description": "(Optional) Additional filters to apply to the search"
},
"max_num_results": {
"type": "integer",
"default": 10,
"description": "(Optional) Maximum number of search results to return (1-50)"
},
"ranking_options": {
"type": "object",
"properties": {
"ranker": {
"type": "string",
"description": "(Optional) Name of the ranking algorithm to use"
},
"score_threshold": {
"type": "number",
"default": 0.0,
"description": "(Optional) Minimum relevance score threshold for results"
}
},
"additionalProperties": false,
"description": "(Optional) Options for ranking and scoring search results"
}
},
"additionalProperties": false,
"required": [
"type",
"vector_store_ids"
],
"title": "OpenAIResponseInputToolFileSearch",
"description": "File search tool configuration for OpenAI response inputs."
},
"OpenAIResponseInputToolFunction": {
"type": "object",
"properties": {
"type": {
"type": "string",
"const": "function",
"default": "function",
"description": "Tool type identifier, always \"function\""
},
"name": {
"type": "string",
"description": "Name of the function that can be called"
},
"description": {
"type": "string",
"description": "(Optional) Description of what the function does"
},
"parameters": {
"type": "object",
"additionalProperties": {
"oneOf": [
{
"type": "null"
},
{
"type": "boolean"
},
{
"type": "number"
},
{
"type": "string"
},
{
"type": "array"
},
{
"type": "object"
}
]
},
"description": "(Optional) JSON schema defining the function's parameters"
},
"strict": {
"type": "boolean",
"description": "(Optional) Whether to enforce strict parameter validation"
}
},
"additionalProperties": false,
"required": [
"type",
"name"
],
"title": "OpenAIResponseInputToolFunction",
"description": "Function tool configuration for OpenAI response inputs."
},
"OpenAIResponseInputToolWebSearch": {
"type": "object",
"properties": {
"type": {
"oneOf": [
{
"type": "string",
"const": "web_search"
},
{
"type": "string",
"const": "web_search_preview"
},
{
"type": "string",
"const": "web_search_preview_2025_03_11"
}
],
"default": "web_search",
"description": "Web search tool type variant to use"
},
"search_context_size": {
"type": "string",
"default": "medium",
"description": "(Optional) Size of search context, must be \"low\", \"medium\", or \"high\""
}
},
"additionalProperties": false,
"required": [
"type"
],
"title": "OpenAIResponseInputToolWebSearch",
"description": "Web search tool configuration for OpenAI response inputs."
},
"OpenAIResponseMCPApprovalRequest": {
"type": "object",
"properties": {
@ -7564,6 +7726,13 @@
"type": "number",
"description": "(Optional) Nucleus sampling parameter used for generation"
},
"tools": {
"type": "array",
"items": {
"$ref": "#/components/schemas/OpenAIResponseTool"
},
"description": "(Optional) An array of tools the model may call while generating a response."
},
"truncation": {
"type": "string",
"description": "(Optional) Truncation strategy applied to the response"
@ -7705,6 +7874,79 @@
"title": "OpenAIResponseText",
"description": "Text response configuration for OpenAI responses."
},
"OpenAIResponseTool": {
"oneOf": [
{
"$ref": "#/components/schemas/OpenAIResponseInputToolWebSearch"
},
{
"$ref": "#/components/schemas/OpenAIResponseInputToolFileSearch"
},
{
"$ref": "#/components/schemas/OpenAIResponseInputToolFunction"
},
{
"$ref": "#/components/schemas/OpenAIResponseToolMCP"
}
],
"discriminator": {
"propertyName": "type",
"mapping": {
"web_search": "#/components/schemas/OpenAIResponseInputToolWebSearch",
"file_search": "#/components/schemas/OpenAIResponseInputToolFileSearch",
"function": "#/components/schemas/OpenAIResponseInputToolFunction",
"mcp": "#/components/schemas/OpenAIResponseToolMCP"
}
}
},
"OpenAIResponseToolMCP": {
"type": "object",
"properties": {
"type": {
"type": "string",
"const": "mcp",
"default": "mcp",
"description": "Tool type identifier, always \"mcp\""
},
"server_label": {
"type": "string",
"description": "Label to identify this MCP server"
},
"allowed_tools": {
"oneOf": [
{
"type": "array",
"items": {
"type": "string"
}
},
{
"type": "object",
"properties": {
"tool_names": {
"type": "array",
"items": {
"type": "string"
},
"description": "(Optional) List of specific tool names that are allowed"
}
},
"additionalProperties": false,
"title": "AllowedToolsFilter",
"description": "Filter configuration for restricting which MCP tools can be used."
}
],
"description": "(Optional) Restriction on which tools can be used from this server"
}
},
"additionalProperties": false,
"required": [
"type",
"server_label"
],
"title": "OpenAIResponseToolMCP",
"description": "Model Context Protocol (MCP) tool configuration for OpenAI response object."
},
"OpenAIResponseUsage": {
"type": "object",
"properties": {
@ -7792,134 +8034,6 @@
}
}
},
"OpenAIResponseInputToolFileSearch": {
"type": "object",
"properties": {
"type": {
"type": "string",
"const": "file_search",
"default": "file_search",
"description": "Tool type identifier, always \"file_search\""
},
"vector_store_ids": {
"type": "array",
"items": {
"type": "string"
},
"description": "List of vector store identifiers to search within"
},
"filters": {
"type": "object",
"additionalProperties": {
"oneOf": [
{
"type": "null"
},
{
"type": "boolean"
},
{
"type": "number"
},
{
"type": "string"
},
{
"type": "array"
},
{
"type": "object"
}
]
},
"description": "(Optional) Additional filters to apply to the search"
},
"max_num_results": {
"type": "integer",
"default": 10,
"description": "(Optional) Maximum number of search results to return (1-50)"
},
"ranking_options": {
"type": "object",
"properties": {
"ranker": {
"type": "string",
"description": "(Optional) Name of the ranking algorithm to use"
},
"score_threshold": {
"type": "number",
"default": 0.0,
"description": "(Optional) Minimum relevance score threshold for results"
}
},
"additionalProperties": false,
"description": "(Optional) Options for ranking and scoring search results"
}
},
"additionalProperties": false,
"required": [
"type",
"vector_store_ids"
],
"title": "OpenAIResponseInputToolFileSearch",
"description": "File search tool configuration for OpenAI response inputs."
},
"OpenAIResponseInputToolFunction": {
"type": "object",
"properties": {
"type": {
"type": "string",
"const": "function",
"default": "function",
"description": "Tool type identifier, always \"function\""
},
"name": {
"type": "string",
"description": "Name of the function that can be called"
},
"description": {
"type": "string",
"description": "(Optional) Description of what the function does"
},
"parameters": {
"type": "object",
"additionalProperties": {
"oneOf": [
{
"type": "null"
},
{
"type": "boolean"
},
{
"type": "number"
},
{
"type": "string"
},
{
"type": "array"
},
{
"type": "object"
}
]
},
"description": "(Optional) JSON schema defining the function's parameters"
},
"strict": {
"type": "boolean",
"description": "(Optional) Whether to enforce strict parameter validation"
}
},
"additionalProperties": false,
"required": [
"type",
"name"
],
"title": "OpenAIResponseInputToolFunction",
"description": "Function tool configuration for OpenAI response inputs."
},
"OpenAIResponseInputToolMCP": {
"type": "object",
"properties": {
@ -8036,40 +8150,6 @@
"title": "OpenAIResponseInputToolMCP",
"description": "Model Context Protocol (MCP) tool configuration for OpenAI response inputs."
},
"OpenAIResponseInputToolWebSearch": {
"type": "object",
"properties": {
"type": {
"oneOf": [
{
"type": "string",
"const": "web_search"
},
{
"type": "string",
"const": "web_search_preview"
},
{
"type": "string",
"const": "web_search_preview_2025_03_11"
}
],
"default": "web_search",
"description": "Web search tool type variant to use"
},
"search_context_size": {
"type": "string",
"default": "medium",
"description": "(Optional) Size of search context, must be \"low\", \"medium\", or \"high\""
}
},
"additionalProperties": false,
"required": [
"type"
],
"title": "OpenAIResponseInputToolWebSearch",
"description": "Web search tool configuration for OpenAI response inputs."
},
"CreateOpenaiResponseRequest": {
"type": "object",
"properties": {
@ -8191,6 +8271,13 @@
"type": "number",
"description": "(Optional) Nucleus sampling parameter used for generation"
},
"tools": {
"type": "array",
"items": {
"$ref": "#/components/schemas/OpenAIResponseTool"
},
"description": "(Optional) An array of tools the model may call while generating a response."
},
"truncation": {
"type": "string",
"description": "(Optional) Truncation strategy applied to the response"

294
docs/static/llama-stack-spec.yaml vendored
View file

@ -5654,6 +5654,122 @@ components:
description: >-
This represents the output of a function call that gets passed back to the
model.
OpenAIResponseInputToolFileSearch:
type: object
properties:
type:
type: string
const: file_search
default: file_search
description: >-
Tool type identifier, always "file_search"
vector_store_ids:
type: array
items:
type: string
description: >-
List of vector store identifiers to search within
filters:
type: object
additionalProperties:
oneOf:
- type: 'null'
- type: boolean
- type: number
- type: string
- type: array
- type: object
description: >-
(Optional) Additional filters to apply to the search
max_num_results:
type: integer
default: 10
description: >-
(Optional) Maximum number of search results to return (1-50)
ranking_options:
type: object
properties:
ranker:
type: string
description: >-
(Optional) Name of the ranking algorithm to use
score_threshold:
type: number
default: 0.0
description: >-
(Optional) Minimum relevance score threshold for results
additionalProperties: false
description: >-
(Optional) Options for ranking and scoring search results
additionalProperties: false
required:
- type
- vector_store_ids
title: OpenAIResponseInputToolFileSearch
description: >-
File search tool configuration for OpenAI response inputs.
OpenAIResponseInputToolFunction:
type: object
properties:
type:
type: string
const: function
default: function
description: Tool type identifier, always "function"
name:
type: string
description: Name of the function that can be called
description:
type: string
description: >-
(Optional) Description of what the function does
parameters:
type: object
additionalProperties:
oneOf:
- type: 'null'
- type: boolean
- type: number
- type: string
- type: array
- type: object
description: >-
(Optional) JSON schema defining the function's parameters
strict:
type: boolean
description: >-
(Optional) Whether to enforce strict parameter validation
additionalProperties: false
required:
- type
- name
title: OpenAIResponseInputToolFunction
description: >-
Function tool configuration for OpenAI response inputs.
OpenAIResponseInputToolWebSearch:
type: object
properties:
type:
oneOf:
- type: string
const: web_search
- type: string
const: web_search_preview
- type: string
const: web_search_preview_2025_03_11
default: web_search
description: Web search tool type variant to use
search_context_size:
type: string
default: medium
description: >-
(Optional) Size of search context, must be "low", "medium", or "high"
additionalProperties: false
required:
- type
title: OpenAIResponseInputToolWebSearch
description: >-
Web search tool configuration for OpenAI response inputs.
OpenAIResponseMCPApprovalRequest:
type: object
properties:
@ -5755,6 +5871,12 @@ components:
type: number
description: >-
(Optional) Nucleus sampling parameter used for generation
tools:
type: array
items:
$ref: '#/components/schemas/OpenAIResponseTool'
description: >-
(Optional) An array of tools the model may call while generating a response.
truncation:
type: string
description: >-
@ -5854,6 +5976,56 @@ components:
title: OpenAIResponseText
description: >-
Text response configuration for OpenAI responses.
OpenAIResponseTool:
oneOf:
- $ref: '#/components/schemas/OpenAIResponseInputToolWebSearch'
- $ref: '#/components/schemas/OpenAIResponseInputToolFileSearch'
- $ref: '#/components/schemas/OpenAIResponseInputToolFunction'
- $ref: '#/components/schemas/OpenAIResponseToolMCP'
discriminator:
propertyName: type
mapping:
web_search: '#/components/schemas/OpenAIResponseInputToolWebSearch'
file_search: '#/components/schemas/OpenAIResponseInputToolFileSearch'
function: '#/components/schemas/OpenAIResponseInputToolFunction'
mcp: '#/components/schemas/OpenAIResponseToolMCP'
OpenAIResponseToolMCP:
type: object
properties:
type:
type: string
const: mcp
default: mcp
description: Tool type identifier, always "mcp"
server_label:
type: string
description: Label to identify this MCP server
allowed_tools:
oneOf:
- type: array
items:
type: string
- type: object
properties:
tool_names:
type: array
items:
type: string
description: >-
(Optional) List of specific tool names that are allowed
additionalProperties: false
title: AllowedToolsFilter
description: >-
Filter configuration for restricting which MCP tools can be used.
description: >-
(Optional) Restriction on which tools can be used from this server
additionalProperties: false
required:
- type
- server_label
title: OpenAIResponseToolMCP
description: >-
Model Context Protocol (MCP) tool configuration for OpenAI response object.
OpenAIResponseUsage:
type: object
properties:
@ -5915,98 +6087,6 @@ components:
file_search: '#/components/schemas/OpenAIResponseInputToolFileSearch'
function: '#/components/schemas/OpenAIResponseInputToolFunction'
mcp: '#/components/schemas/OpenAIResponseInputToolMCP'
OpenAIResponseInputToolFileSearch:
type: object
properties:
type:
type: string
const: file_search
default: file_search
description: >-
Tool type identifier, always "file_search"
vector_store_ids:
type: array
items:
type: string
description: >-
List of vector store identifiers to search within
filters:
type: object
additionalProperties:
oneOf:
- type: 'null'
- type: boolean
- type: number
- type: string
- type: array
- type: object
description: >-
(Optional) Additional filters to apply to the search
max_num_results:
type: integer
default: 10
description: >-
(Optional) Maximum number of search results to return (1-50)
ranking_options:
type: object
properties:
ranker:
type: string
description: >-
(Optional) Name of the ranking algorithm to use
score_threshold:
type: number
default: 0.0
description: >-
(Optional) Minimum relevance score threshold for results
additionalProperties: false
description: >-
(Optional) Options for ranking and scoring search results
additionalProperties: false
required:
- type
- vector_store_ids
title: OpenAIResponseInputToolFileSearch
description: >-
File search tool configuration for OpenAI response inputs.
OpenAIResponseInputToolFunction:
type: object
properties:
type:
type: string
const: function
default: function
description: Tool type identifier, always "function"
name:
type: string
description: Name of the function that can be called
description:
type: string
description: >-
(Optional) Description of what the function does
parameters:
type: object
additionalProperties:
oneOf:
- type: 'null'
- type: boolean
- type: number
- type: string
- type: array
- type: object
description: >-
(Optional) JSON schema defining the function's parameters
strict:
type: boolean
description: >-
(Optional) Whether to enforce strict parameter validation
additionalProperties: false
required:
- type
- name
title: OpenAIResponseInputToolFunction
description: >-
Function tool configuration for OpenAI response inputs.
OpenAIResponseInputToolMCP:
type: object
properties:
@ -6088,30 +6168,6 @@ components:
title: OpenAIResponseInputToolMCP
description: >-
Model Context Protocol (MCP) tool configuration for OpenAI response inputs.
OpenAIResponseInputToolWebSearch:
type: object
properties:
type:
oneOf:
- type: string
const: web_search
- type: string
const: web_search_preview
- type: string
const: web_search_preview_2025_03_11
default: web_search
description: Web search tool type variant to use
search_context_size:
type: string
default: medium
description: >-
(Optional) Size of search context, must be "low", "medium", or "high"
additionalProperties: false
required:
- type
title: OpenAIResponseInputToolWebSearch
description: >-
Web search tool configuration for OpenAI response inputs.
CreateOpenaiResponseRequest:
type: object
properties:
@ -6212,6 +6268,12 @@ components:
type: number
description: >-
(Optional) Nucleus sampling parameter used for generation
tools:
type: array
items:
$ref: '#/components/schemas/OpenAIResponseTool'
description: >-
(Optional) An array of tools the model may call while generating a response.
truncation:
type: string
description: >-

411
docs/static/stainless-llama-stack-spec.html vendored
View file

@ -9454,6 +9454,168 @@
"title": "OpenAIResponseInputFunctionToolCallOutput",
"description": "This represents the output of a function call that gets passed back to the model."
},
"OpenAIResponseInputToolFileSearch": {
"type": "object",
"properties": {
"type": {
"type": "string",
"const": "file_search",
"default": "file_search",
"description": "Tool type identifier, always \"file_search\""
},
"vector_store_ids": {
"type": "array",
"items": {
"type": "string"
},
"description": "List of vector store identifiers to search within"
},
"filters": {
"type": "object",
"additionalProperties": {
"oneOf": [
{
"type": "null"
},
{
"type": "boolean"
},
{
"type": "number"
},
{
"type": "string"
},
{
"type": "array"
},
{
"type": "object"
}
]
},
"description": "(Optional) Additional filters to apply to the search"
},
"max_num_results": {
"type": "integer",
"default": 10,
"description": "(Optional) Maximum number of search results to return (1-50)"
},
"ranking_options": {
"type": "object",
"properties": {
"ranker": {
"type": "string",
"description": "(Optional) Name of the ranking algorithm to use"
},
"score_threshold": {
"type": "number",
"default": 0.0,
"description": "(Optional) Minimum relevance score threshold for results"
}
},
"additionalProperties": false,
"description": "(Optional) Options for ranking and scoring search results"
}
},
"additionalProperties": false,
"required": [
"type",
"vector_store_ids"
],
"title": "OpenAIResponseInputToolFileSearch",
"description": "File search tool configuration for OpenAI response inputs."
},
"OpenAIResponseInputToolFunction": {
"type": "object",
"properties": {
"type": {
"type": "string",
"const": "function",
"default": "function",
"description": "Tool type identifier, always \"function\""
},
"name": {
"type": "string",
"description": "Name of the function that can be called"
},
"description": {
"type": "string",
"description": "(Optional) Description of what the function does"
},
"parameters": {
"type": "object",
"additionalProperties": {
"oneOf": [
{
"type": "null"
},
{
"type": "boolean"
},
{
"type": "number"
},
{
"type": "string"
},
{
"type": "array"
},
{
"type": "object"
}
]
},
"description": "(Optional) JSON schema defining the function's parameters"
},
"strict": {
"type": "boolean",
"description": "(Optional) Whether to enforce strict parameter validation"
}
},
"additionalProperties": false,
"required": [
"type",
"name"
],
"title": "OpenAIResponseInputToolFunction",
"description": "Function tool configuration for OpenAI response inputs."
},
"OpenAIResponseInputToolWebSearch": {
"type": "object",
"properties": {
"type": {
"oneOf": [
{
"type": "string",
"const": "web_search"
},
{
"type": "string",
"const": "web_search_preview"
},
{
"type": "string",
"const": "web_search_preview_2025_03_11"
}
],
"default": "web_search",
"description": "Web search tool type variant to use"
},
"search_context_size": {
"type": "string",
"default": "medium",
"description": "(Optional) Size of search context, must be \"low\", \"medium\", or \"high\""
}
},
"additionalProperties": false,
"required": [
"type"
],
"title": "OpenAIResponseInputToolWebSearch",
"description": "Web search tool configuration for OpenAI response inputs."
},
"OpenAIResponseMCPApprovalRequest": {
"type": "object",
"properties": {
@ -9573,6 +9735,13 @@
"type": "number",
"description": "(Optional) Nucleus sampling parameter used for generation"
},
"tools": {
"type": "array",
"items": {
"$ref": "#/components/schemas/OpenAIResponseTool"
},
"description": "(Optional) An array of tools the model may call while generating a response."
},
"truncation": {
"type": "string",
"description": "(Optional) Truncation strategy applied to the response"
@ -9714,6 +9883,79 @@
"title": "OpenAIResponseText",
"description": "Text response configuration for OpenAI responses."
},
"OpenAIResponseTool": {
"oneOf": [
{
"$ref": "#/components/schemas/OpenAIResponseInputToolWebSearch"
},
{
"$ref": "#/components/schemas/OpenAIResponseInputToolFileSearch"
},
{
"$ref": "#/components/schemas/OpenAIResponseInputToolFunction"
},
{
"$ref": "#/components/schemas/OpenAIResponseToolMCP"
}
],
"discriminator": {
"propertyName": "type",
"mapping": {
"web_search": "#/components/schemas/OpenAIResponseInputToolWebSearch",
"file_search": "#/components/schemas/OpenAIResponseInputToolFileSearch",
"function": "#/components/schemas/OpenAIResponseInputToolFunction",
"mcp": "#/components/schemas/OpenAIResponseToolMCP"
}
}
},
"OpenAIResponseToolMCP": {
"type": "object",
"properties": {
"type": {
"type": "string",
"const": "mcp",
"default": "mcp",
"description": "Tool type identifier, always \"mcp\""
},
"server_label": {
"type": "string",
"description": "Label to identify this MCP server"
},
"allowed_tools": {
"oneOf": [
{
"type": "array",
"items": {
"type": "string"
}
},
{
"type": "object",
"properties": {
"tool_names": {
"type": "array",
"items": {
"type": "string"
},
"description": "(Optional) List of specific tool names that are allowed"
}
},
"additionalProperties": false,
"title": "AllowedToolsFilter",
"description": "Filter configuration for restricting which MCP tools can be used."
}
],
"description": "(Optional) Restriction on which tools can be used from this server"
}
},
"additionalProperties": false,
"required": [
"type",
"server_label"
],
"title": "OpenAIResponseToolMCP",
"description": "Model Context Protocol (MCP) tool configuration for OpenAI response object."
},
"OpenAIResponseUsage": {
"type": "object",
"properties": {
@ -9801,134 +10043,6 @@
}
}
},
"OpenAIResponseInputToolFileSearch": {
"type": "object",
"properties": {
"type": {
"type": "string",
"const": "file_search",
"default": "file_search",
"description": "Tool type identifier, always \"file_search\""
},
"vector_store_ids": {
"type": "array",
"items": {
"type": "string"
},
"description": "List of vector store identifiers to search within"
},
"filters": {
"type": "object",
"additionalProperties": {
"oneOf": [
{
"type": "null"
},
{
"type": "boolean"
},
{
"type": "number"
},
{
"type": "string"
},
{
"type": "array"
},
{
"type": "object"
}
]
},
"description": "(Optional) Additional filters to apply to the search"
},
"max_num_results": {
"type": "integer",
"default": 10,
"description": "(Optional) Maximum number of search results to return (1-50)"
},
"ranking_options": {
"type": "object",
"properties": {
"ranker": {
"type": "string",
"description": "(Optional) Name of the ranking algorithm to use"
},
"score_threshold": {
"type": "number",
"default": 0.0,
"description": "(Optional) Minimum relevance score threshold for results"
}
},
"additionalProperties": false,
"description": "(Optional) Options for ranking and scoring search results"
}
},
"additionalProperties": false,
"required": [
"type",
"vector_store_ids"
],
"title": "OpenAIResponseInputToolFileSearch",
"description": "File search tool configuration for OpenAI response inputs."
},
"OpenAIResponseInputToolFunction": {
"type": "object",
"properties": {
"type": {
"type": "string",
"const": "function",
"default": "function",
"description": "Tool type identifier, always \"function\""
},
"name": {
"type": "string",
"description": "Name of the function that can be called"
},
"description": {
"type": "string",
"description": "(Optional) Description of what the function does"
},
"parameters": {
"type": "object",
"additionalProperties": {
"oneOf": [
{
"type": "null"
},
{
"type": "boolean"
},
{
"type": "number"
},
{
"type": "string"
},
{
"type": "array"
},
{
"type": "object"
}
]
},
"description": "(Optional) JSON schema defining the function's parameters"
},
"strict": {
"type": "boolean",
"description": "(Optional) Whether to enforce strict parameter validation"
}
},
"additionalProperties": false,
"required": [
"type",
"name"
],
"title": "OpenAIResponseInputToolFunction",
"description": "Function tool configuration for OpenAI response inputs."
},
"OpenAIResponseInputToolMCP": {
"type": "object",
"properties": {
@ -10045,40 +10159,6 @@
"title": "OpenAIResponseInputToolMCP",
"description": "Model Context Protocol (MCP) tool configuration for OpenAI response inputs."
},
"OpenAIResponseInputToolWebSearch": {
"type": "object",
"properties": {
"type": {
"oneOf": [
{
"type": "string",
"const": "web_search"
},
{
"type": "string",
"const": "web_search_preview"
},
{
"type": "string",
"const": "web_search_preview_2025_03_11"
}
],
"default": "web_search",
"description": "Web search tool type variant to use"
},
"search_context_size": {
"type": "string",
"default": "medium",
"description": "(Optional) Size of search context, must be \"low\", \"medium\", or \"high\""
}
},
"additionalProperties": false,
"required": [
"type"
],
"title": "OpenAIResponseInputToolWebSearch",
"description": "Web search tool configuration for OpenAI response inputs."
},
"CreateOpenaiResponseRequest": {
"type": "object",
"properties": {
@ -10200,6 +10280,13 @@
"type": "number",
"description": "(Optional) Nucleus sampling parameter used for generation"
},
"tools": {
"type": "array",
"items": {
"$ref": "#/components/schemas/OpenAIResponseTool"
},
"description": "(Optional) An array of tools the model may call while generating a response."
},
"truncation": {
"type": "string",
"description": "(Optional) Truncation strategy applied to the response"

294
docs/static/stainless-llama-stack-spec.yaml vendored
View file

@ -7099,6 +7099,122 @@ components:
description: >-
This represents the output of a function call that gets passed back to the
model.
OpenAIResponseInputToolFileSearch:
type: object
properties:
type:
type: string
const: file_search
default: file_search
description: >-
Tool type identifier, always "file_search"
vector_store_ids:
type: array
items:
type: string
description: >-
List of vector store identifiers to search within
filters:
type: object
additionalProperties:
oneOf:
- type: 'null'
- type: boolean
- type: number
- type: string
- type: array
- type: object
description: >-
(Optional) Additional filters to apply to the search
max_num_results:
type: integer
default: 10
description: >-
(Optional) Maximum number of search results to return (1-50)
ranking_options:
type: object
properties:
ranker:
type: string
description: >-
(Optional) Name of the ranking algorithm to use
score_threshold:
type: number
default: 0.0
description: >-
(Optional) Minimum relevance score threshold for results
additionalProperties: false
description: >-
(Optional) Options for ranking and scoring search results
additionalProperties: false
required:
- type
- vector_store_ids
title: OpenAIResponseInputToolFileSearch
description: >-
File search tool configuration for OpenAI response inputs.
OpenAIResponseInputToolFunction:
type: object
properties:
type:
type: string
const: function
default: function
description: Tool type identifier, always "function"
name:
type: string
description: Name of the function that can be called
description:
type: string
description: >-
(Optional) Description of what the function does
parameters:
type: object
additionalProperties:
oneOf:
- type: 'null'
- type: boolean
- type: number
- type: string
- type: array
- type: object
description: >-
(Optional) JSON schema defining the function's parameters
strict:
type: boolean
description: >-
(Optional) Whether to enforce strict parameter validation
additionalProperties: false
required:
- type
- name
title: OpenAIResponseInputToolFunction
description: >-
Function tool configuration for OpenAI response inputs.
OpenAIResponseInputToolWebSearch:
type: object
properties:
type:
oneOf:
- type: string
const: web_search
- type: string
const: web_search_preview
- type: string
const: web_search_preview_2025_03_11
default: web_search
description: Web search tool type variant to use
search_context_size:
type: string
default: medium
description: >-
(Optional) Size of search context, must be "low", "medium", or "high"
additionalProperties: false
required:
- type
title: OpenAIResponseInputToolWebSearch
description: >-
Web search tool configuration for OpenAI response inputs.
OpenAIResponseMCPApprovalRequest:
type: object
properties:
@ -7200,6 +7316,12 @@ components:
type: number
description: >-
(Optional) Nucleus sampling parameter used for generation
tools:
type: array
items:
$ref: '#/components/schemas/OpenAIResponseTool'
description: >-
(Optional) An array of tools the model may call while generating a response.
truncation:
type: string
description: >-
@ -7299,6 +7421,56 @@ components:
title: OpenAIResponseText
description: >-
Text response configuration for OpenAI responses.
OpenAIResponseTool:
oneOf:
- $ref: '#/components/schemas/OpenAIResponseInputToolWebSearch'
- $ref: '#/components/schemas/OpenAIResponseInputToolFileSearch'
- $ref: '#/components/schemas/OpenAIResponseInputToolFunction'
- $ref: '#/components/schemas/OpenAIResponseToolMCP'
discriminator:
propertyName: type
mapping:
web_search: '#/components/schemas/OpenAIResponseInputToolWebSearch'
file_search: '#/components/schemas/OpenAIResponseInputToolFileSearch'
function: '#/components/schemas/OpenAIResponseInputToolFunction'
mcp: '#/components/schemas/OpenAIResponseToolMCP'
OpenAIResponseToolMCP:
type: object
properties:
type:
type: string
const: mcp
default: mcp
description: Tool type identifier, always "mcp"
server_label:
type: string
description: Label to identify this MCP server
allowed_tools:
oneOf:
- type: array
items:
type: string
- type: object
properties:
tool_names:
type: array
items:
type: string
description: >-
(Optional) List of specific tool names that are allowed
additionalProperties: false
title: AllowedToolsFilter
description: >-
Filter configuration for restricting which MCP tools can be used.
description: >-
(Optional) Restriction on which tools can be used from this server
additionalProperties: false
required:
- type
- server_label
title: OpenAIResponseToolMCP
description: >-
Model Context Protocol (MCP) tool configuration for OpenAI response object.
OpenAIResponseUsage:
type: object
properties:
@ -7360,98 +7532,6 @@ components:
file_search: '#/components/schemas/OpenAIResponseInputToolFileSearch'
function: '#/components/schemas/OpenAIResponseInputToolFunction'
mcp: '#/components/schemas/OpenAIResponseInputToolMCP'
OpenAIResponseInputToolFileSearch:
type: object
properties:
type:
type: string
const: file_search
default: file_search
description: >-
Tool type identifier, always "file_search"
vector_store_ids:
type: array
items:
type: string
description: >-
List of vector store identifiers to search within
filters:
type: object
additionalProperties:
oneOf:
- type: 'null'
- type: boolean
- type: number
- type: string
- type: array
- type: object
description: >-
(Optional) Additional filters to apply to the search
max_num_results:
type: integer
default: 10
description: >-
(Optional) Maximum number of search results to return (1-50)
ranking_options:
type: object
properties:
ranker:
type: string
description: >-
(Optional) Name of the ranking algorithm to use
score_threshold:
type: number
default: 0.0
description: >-
(Optional) Minimum relevance score threshold for results
additionalProperties: false
description: >-
(Optional) Options for ranking and scoring search results
additionalProperties: false
required:
- type
- vector_store_ids
title: OpenAIResponseInputToolFileSearch
description: >-
File search tool configuration for OpenAI response inputs.
OpenAIResponseInputToolFunction:
type: object
properties:
type:
type: string
const: function
default: function
description: Tool type identifier, always "function"
name:
type: string
description: Name of the function that can be called
description:
type: string
description: >-
(Optional) Description of what the function does
parameters:
type: object
additionalProperties:
oneOf:
- type: 'null'
- type: boolean
- type: number
- type: string
- type: array
- type: object
description: >-
(Optional) JSON schema defining the function's parameters
strict:
type: boolean
description: >-
(Optional) Whether to enforce strict parameter validation
additionalProperties: false
required:
- type
- name
title: OpenAIResponseInputToolFunction
description: >-
Function tool configuration for OpenAI response inputs.
OpenAIResponseInputToolMCP:
type: object
properties:
@ -7533,30 +7613,6 @@ components:
title: OpenAIResponseInputToolMCP
description: >-
Model Context Protocol (MCP) tool configuration for OpenAI response inputs.
OpenAIResponseInputToolWebSearch:
type: object
properties:
type:
oneOf:
- type: string
const: web_search
- type: string
const: web_search_preview
- type: string
const: web_search_preview_2025_03_11
default: web_search
description: Web search tool type variant to use
search_context_size:
type: string
default: medium
description: >-
(Optional) Size of search context, must be "low", "medium", or "high"
additionalProperties: false
required:
- type
title: OpenAIResponseInputToolWebSearch
description: >-
Web search tool configuration for OpenAI response inputs.
CreateOpenaiResponseRequest:
type: object
properties:
@ -7657,6 +7713,12 @@ components:
type: number
description: >-
(Optional) Nucleus sampling parameter used for generation
tools:
type: array
items:
$ref: '#/components/schemas/OpenAIResponseTool'
description: >-
(Optional) An array of tools the model may call while generating a response.
truncation:
type: string
description: >-

242
llama_stack/apis/agents/openai_responses.py
View file

@ -346,6 +346,138 @@ class OpenAIResponseText(BaseModel):
format: OpenAIResponseTextFormat | None = None
# Must match type Literals of OpenAIResponseInputToolWebSearch below
WebSearchToolTypes = ["web_search", "web_search_preview", "web_search_preview_2025_03_11"]
@json_schema_type
class OpenAIResponseInputToolWebSearch(BaseModel):
"""Web search tool configuration for OpenAI response inputs.
:param type: Web search tool type variant to use
:param search_context_size: (Optional) Size of search context, must be "low", "medium", or "high"
"""
# Must match values of WebSearchToolTypes above
type: Literal["web_search"] | Literal["web_search_preview"] | Literal["web_search_preview_2025_03_11"] = (
"web_search"
)
# TODO: actually use search_context_size somewhere...
search_context_size: str | None = Field(default="medium", pattern="^low|medium|high$")
# TODO: add user_location
@json_schema_type
class OpenAIResponseInputToolFunction(BaseModel):
"""Function tool configuration for OpenAI response inputs.
:param type: Tool type identifier, always "function"
:param name: Name of the function that can be called
:param description: (Optional) Description of what the function does
:param parameters: (Optional) JSON schema defining the function's parameters
:param strict: (Optional) Whether to enforce strict parameter validation
"""
type: Literal["function"] = "function"
name: str
description: str | None = None
parameters: dict[str, Any] | None
strict: bool | None = None
@json_schema_type
class OpenAIResponseInputToolFileSearch(BaseModel):
"""File search tool configuration for OpenAI response inputs.
:param type: Tool type identifier, always "file_search"
:param vector_store_ids: List of vector store identifiers to search within
:param filters: (Optional) Additional filters to apply to the search
:param max_num_results: (Optional) Maximum number of search results to return (1-50)
:param ranking_options: (Optional) Options for ranking and scoring search results
"""
type: Literal["file_search"] = "file_search"
vector_store_ids: list[str]
filters: dict[str, Any] | None = None
max_num_results: int | None = Field(default=10, ge=1, le=50)
ranking_options: FileSearchRankingOptions | None = None
class ApprovalFilter(BaseModel):
"""Filter configuration for MCP tool approval requirements.
:param always: (Optional) List of tool names that always require approval
:param never: (Optional) List of tool names that never require approval
"""
always: list[str] | None = None
never: list[str] | None = None
class AllowedToolsFilter(BaseModel):
"""Filter configuration for restricting which MCP tools can be used.
:param tool_names: (Optional) List of specific tool names that are allowed
"""
tool_names: list[str] | None = None
@json_schema_type
class OpenAIResponseInputToolMCP(BaseModel):
"""Model Context Protocol (MCP) tool configuration for OpenAI response inputs.
:param type: Tool type identifier, always "mcp"
:param server_label: Label to identify this MCP server
:param server_url: URL endpoint of the MCP server
:param headers: (Optional) HTTP headers to include when connecting to the server
:param require_approval: Approval requirement for tool calls ("always", "never", or filter)
:param allowed_tools: (Optional) Restriction on which tools can be used from this server
"""
type: Literal["mcp"] = "mcp"
server_label: str
server_url: str
headers: dict[str, Any] | None = None
require_approval: Literal["always"] | Literal["never"] | ApprovalFilter = "never"
allowed_tools: list[str] | AllowedToolsFilter | None = None
OpenAIResponseInputTool = Annotated[
OpenAIResponseInputToolWebSearch
| OpenAIResponseInputToolFileSearch
| OpenAIResponseInputToolFunction
| OpenAIResponseInputToolMCP,
Field(discriminator="type"),
]
register_schema(OpenAIResponseInputTool, name="OpenAIResponseInputTool")
@json_schema_type
class OpenAIResponseToolMCP(BaseModel):
"""Model Context Protocol (MCP) tool configuration for OpenAI response object.
:param type: Tool type identifier, always "mcp"
:param server_label: Label to identify this MCP server
:param allowed_tools: (Optional) Restriction on which tools can be used from this server
"""
type: Literal["mcp"] = "mcp"
server_label: str
allowed_tools: list[str] | AllowedToolsFilter | None = None
OpenAIResponseTool = Annotated[
OpenAIResponseInputToolWebSearch
| OpenAIResponseInputToolFileSearch
| OpenAIResponseInputToolFunction
| OpenAIResponseToolMCP, # The only type that differes from that in the inputs is the MCP tool
Field(discriminator="type"),
]
register_schema(OpenAIResponseTool, name="OpenAIResponseTool")
class OpenAIResponseUsageOutputTokensDetails(BaseModel):
"""Token details for output tokens in OpenAI response usage.
@ -398,6 +530,7 @@ class OpenAIResponseObject(BaseModel):
:param temperature: (Optional) Sampling temperature used for generation
:param text: Text formatting configuration for the response
:param top_p: (Optional) Nucleus sampling parameter used for generation
:param tools: (Optional) An array of tools the model may call while generating a response.
:param truncation: (Optional) Truncation strategy applied to the response
:param usage: (Optional) Token usage information for the response
"""
@ -416,6 +549,7 @@ class OpenAIResponseObject(BaseModel):
# before the field was added. New responses will have this set always.
text: OpenAIResponseText = OpenAIResponseText(format=OpenAIResponseTextFormat(type="text"))
top_p: float | None = None
tools: list[OpenAIResponseTool] | None = None
truncation: str | None = None
usage: OpenAIResponseUsage | None = None
@ -878,114 +1012,6 @@ OpenAIResponseInput = Annotated[
register_schema(OpenAIResponseInput, name="OpenAIResponseInput")
# Must match type Literals of OpenAIResponseInputToolWebSearch below
WebSearchToolTypes = ["web_search", "web_search_preview", "web_search_preview_2025_03_11"]
@json_schema_type
class OpenAIResponseInputToolWebSearch(BaseModel):
"""Web search tool configuration for OpenAI response inputs.
:param type: Web search tool type variant to use
:param search_context_size: (Optional) Size of search context, must be "low", "medium", or "high"
"""
# Must match values of WebSearchToolTypes above
type: Literal["web_search"] | Literal["web_search_preview"] | Literal["web_search_preview_2025_03_11"] = (
"web_search"
)
# TODO: actually use search_context_size somewhere...
search_context_size: str | None = Field(default="medium", pattern="^low|medium|high$")
# TODO: add user_location
@json_schema_type
class OpenAIResponseInputToolFunction(BaseModel):
"""Function tool configuration for OpenAI response inputs.
:param type: Tool type identifier, always "function"
:param name: Name of the function that can be called
:param description: (Optional) Description of what the function does
:param parameters: (Optional) JSON schema defining the function's parameters
:param strict: (Optional) Whether to enforce strict parameter validation
"""
type: Literal["function"] = "function"
name: str
description: str | None = None
parameters: dict[str, Any] | None
strict: bool | None = None
@json_schema_type
class OpenAIResponseInputToolFileSearch(BaseModel):
"""File search tool configuration for OpenAI response inputs.
:param type: Tool type identifier, always "file_search"
:param vector_store_ids: List of vector store identifiers to search within
:param filters: (Optional) Additional filters to apply to the search
:param max_num_results: (Optional) Maximum number of search results to return (1-50)
:param ranking_options: (Optional) Options for ranking and scoring search results
"""
type: Literal["file_search"] = "file_search"
vector_store_ids: list[str]
filters: dict[str, Any] | None = None
max_num_results: int | None = Field(default=10, ge=1, le=50)
ranking_options: FileSearchRankingOptions | None = None
class ApprovalFilter(BaseModel):
"""Filter configuration for MCP tool approval requirements.
:param always: (Optional) List of tool names that always require approval
:param never: (Optional) List of tool names that never require approval
"""
always: list[str] | None = None
never: list[str] | None = None
class AllowedToolsFilter(BaseModel):
"""Filter configuration for restricting which MCP tools can be used.
:param tool_names: (Optional) List of specific tool names that are allowed
"""
tool_names: list[str] | None = None
@json_schema_type
class OpenAIResponseInputToolMCP(BaseModel):
"""Model Context Protocol (MCP) tool configuration for OpenAI response inputs.
:param type: Tool type identifier, always "mcp"
:param server_label: Label to identify this MCP server
:param server_url: URL endpoint of the MCP server
:param headers: (Optional) HTTP headers to include when connecting to the server
:param require_approval: Approval requirement for tool calls ("always", "never", or filter)
:param allowed_tools: (Optional) Restriction on which tools can be used from this server
"""
type: Literal["mcp"] = "mcp"
server_label: str
server_url: str
headers: dict[str, Any] | None = None
require_approval: Literal["always"] | Literal["never"] | ApprovalFilter = "never"
allowed_tools: list[str] | AllowedToolsFilter | None = None
OpenAIResponseInputTool = Annotated[
OpenAIResponseInputToolWebSearch
| OpenAIResponseInputToolFileSearch
| OpenAIResponseInputToolFunction
| OpenAIResponseInputToolMCP,
Field(discriminator="type"),
]
register_schema(OpenAIResponseInputTool, name="OpenAIResponseInputTool")
class ListOpenAIResponseInputItem(BaseModel):
"""List container for OpenAI response input items.

15
llama_stack/providers/inline/agents/meta_reference/responses/openai_responses.py
View file

@ -39,7 +39,7 @@ from llama_stack.providers.utils.responses.responses_store import (
from .streaming import StreamingResponseOrchestrator
from .tool_executor import ToolExecutor
from .types import ChatCompletionContext
from .types import ChatCompletionContext, ToolContext
from .utils import (
convert_response_input_to_chat_messages,
convert_response_text_to_chat_response_format,
@ -91,13 +91,15 @@ class OpenAIResponsesImpl:
async def _process_input_with_previous_response(
self,
input: str | list[OpenAIResponseInput],
tools: list[OpenAIResponseInputTool] | None,
previous_response_id: str | None,
) -> tuple[str | list[OpenAIResponseInput], list[OpenAIMessageParam]]:
"""Process input with optional previous response context.
Returns:
tuple: (all_input for storage, messages for chat completion)
tuple: (all_input for storage, messages for chat completion, tool context)
"""
tool_context = ToolContext(tools)
if previous_response_id:
previous_response: _OpenAIResponseObjectWithInputAndMessages = (
await self.responses_store.get_response_object(previous_response_id)
@ -113,11 +115,13 @@ class OpenAIResponsesImpl:
else:
# Backward compatibility: reconstruct from inputs
messages = await convert_response_input_to_chat_messages(all_input)
tool_context.recover_tools_from_previous_response(previous_response)
else:
all_input = input
messages = await convert_response_input_to_chat_messages(input)
return all_input, messages
return all_input, messages, tool_context
async def _prepend_instructions(self, messages, instructions):
if instructions:
@ -273,7 +277,9 @@ class OpenAIResponsesImpl:
max_infer_iters: int | None = 10,
) -> AsyncIterator[OpenAIResponseObjectStream]:
# Input preprocessing
all_input, messages = await self._process_input_with_previous_response(input, previous_response_id)
all_input, messages, tool_context = await self._process_input_with_previous_response(
input, tools, previous_response_id
)
await self._prepend_instructions(messages, instructions)
# Structured outputs
@ -285,6 +291,7 @@ class OpenAIResponsesImpl:
response_tools=tools,
temperature=temperature,
response_format=response_format,
tool_context=tool_context,
inputs=all_input,
)

111
llama_stack/providers/inline/agents/meta_reference/responses/streaming.py
View file

@ -99,7 +99,7 @@ class StreamingResponseOrchestrator:
self.tool_executor = tool_executor
self.sequence_number = 0
# Store MCP tool mapping that gets built during tool processing
self.mcp_tool_to_server: dict[str, OpenAIResponseInputToolMCP] = {}
self.mcp_tool_to_server: dict[str, OpenAIResponseInputToolMCP] = ctx.tool_context.previous_tools or {}
# Track final messages after all tool executions
self.final_messages: list[OpenAIMessageParam] = []
# mapping for annotations
@ -129,6 +129,7 @@ class StreamingResponseOrchestrator:
status=status,
output=self._clone_outputs(outputs),
text=self.text,
tools=self.ctx.available_tools(),
error=error,
)
@ -146,10 +147,8 @@ class StreamingResponseOrchestrator:
sequence_number=self.sequence_number,
)
# Process all tools (including MCP tools) and emit streaming events
if self.ctx.response_tools:
async for stream_event in self._process_tools(self.ctx.response_tools, output_messages):
yield stream_event
async for stream_event in self._process_tools(output_messages):
yield stream_event
n_iter = 0
messages = self.ctx.messages.copy()
@ -590,7 +589,7 @@ class StreamingResponseOrchestrator:
sequence_number=self.sequence_number,
)
async def _process_tools(
async def _process_new_tools(
self, tools: list[OpenAIResponseInputTool], output_messages: list[OpenAIResponseOutput]
) -> AsyncIterator[OpenAIResponseObjectStream]:
"""Process all tools and emit appropriate streaming events."""
@ -645,7 +644,6 @@ class StreamingResponseOrchestrator:
yield OpenAIResponseObjectStreamResponseMcpListToolsInProgress(
sequence_number=self.sequence_number,
)
try:
# Parse allowed/never allowed tools
always_allowed = None
@ -707,39 +705,26 @@ class StreamingResponseOrchestrator:
},
)
)
# Add the MCP list message to output
output_messages.append(mcp_list_message)
# Emit output_item.added for the MCP list tools message
self.sequence_number += 1
yield OpenAIResponseObjectStreamResponseOutputItemAdded(
response_id=self.response_id,
item=mcp_list_message,
output_index=len(output_messages) - 1,
sequence_number=self.sequence_number,
)
# Emit mcp_list_tools.completed
self.sequence_number += 1
yield OpenAIResponseObjectStreamResponseMcpListToolsCompleted(
sequence_number=self.sequence_number,
)
# Emit output_item.done for the MCP list tools message
self.sequence_number += 1
yield OpenAIResponseObjectStreamResponseOutputItemDone(
response_id=self.response_id,
item=mcp_list_message,
output_index=len(output_messages) - 1,
sequence_number=self.sequence_number,
)
async for stream_event in self._add_mcp_list_tools(mcp_list_message, output_messages):
yield stream_event
except Exception as e:
# TODO: Emit mcp_list_tools.failed event if needed
logger.exception(f"Failed to list MCP tools from {mcp_tool.server_url}: {e}")
raise
async def _process_tools(
self, output_messages: list[OpenAIResponseOutput]
) -> AsyncIterator[OpenAIResponseObjectStream]:
# Handle all mcp tool lists from previous response that are still valid:
for tool in self.ctx.tool_context.previous_tool_listings:
async for evt in self._reuse_mcp_list_tools(tool, output_messages):
yield evt
# Process all remaining tools (including MCP tools) and emit streaming events
if self.ctx.tool_context.tools_to_process:
async for stream_event in self._process_new_tools(self.ctx.tool_context.tools_to_process, output_messages):
yield stream_event
def _approval_required(self, tool_name: str) -> bool:
if tool_name not in self.mcp_tool_to_server:
return False
@ -774,7 +759,6 @@ class StreamingResponseOrchestrator:
output_index=len(output_messages) - 1,
sequence_number=self.sequence_number,
)
self.sequence_number += 1
yield OpenAIResponseObjectStreamResponseOutputItemDone(
response_id=self.response_id,
@ -782,3 +766,60 @@ class StreamingResponseOrchestrator:
output_index=len(output_messages) - 1,
sequence_number=self.sequence_number,
)
async def _add_mcp_list_tools(
self, mcp_list_message: OpenAIResponseOutputMessageMCPListTools, output_messages: list[OpenAIResponseOutput]
) -> AsyncIterator[OpenAIResponseObjectStream]:
# Add the MCP list message to output
output_messages.append(mcp_list_message)
# Emit output_item.added for the MCP list tools message
self.sequence_number += 1
yield OpenAIResponseObjectStreamResponseOutputItemAdded(
response_id=self.response_id,
item=mcp_list_message,
output_index=len(output_messages) - 1,
sequence_number=self.sequence_number,
)
# Emit mcp_list_tools.completed
self.sequence_number += 1
yield OpenAIResponseObjectStreamResponseMcpListToolsCompleted(
sequence_number=self.sequence_number,
)
# Emit output_item.done for the MCP list tools message
self.sequence_number += 1
yield OpenAIResponseObjectStreamResponseOutputItemDone(
response_id=self.response_id,
item=mcp_list_message,
output_index=len(output_messages) - 1,
sequence_number=self.sequence_number,
)
async def _reuse_mcp_list_tools(
self, original: OpenAIResponseOutputMessageMCPListTools, output_messages: list[OpenAIResponseOutput]
) -> AsyncIterator[OpenAIResponseObjectStream]:
for t in original.tools:
from llama_stack.models.llama.datatypes import ToolDefinition
from llama_stack.providers.utils.inference.openai_compat import convert_tooldef_to_openai_tool
# convert from input_schema to map of ToolParamDefinitions...
tool_def = ToolDefinition(
tool_name=t.name,
description=t.description,
input_schema=t.input_schema,
)
# ...then can convert that to openai completions tool
openai_tool = convert_tooldef_to_openai_tool(tool_def)
if self.ctx.chat_tools is None:
self.ctx.chat_tools = []
self.ctx.chat_tools.append(openai_tool)
mcp_list_message = OpenAIResponseOutputMessageMCPListTools(
id=f"mcp_list_{uuid.uuid4()}",
server_label=original.server_label,
tools=original.tools,
)
async for stream_event in self._add_mcp_list_tools(mcp_list_message, output_messages):
yield stream_event

96
llama_stack/providers/inline/agents/meta_reference/responses/types.py
View file

@ -12,10 +12,18 @@ from pydantic import BaseModel
from llama_stack.apis.agents.openai_responses import (
OpenAIResponseInput,
OpenAIResponseInputTool,
OpenAIResponseInputToolFileSearch,
OpenAIResponseInputToolFunction,
OpenAIResponseInputToolMCP,
OpenAIResponseInputToolWebSearch,
OpenAIResponseMCPApprovalRequest,
OpenAIResponseMCPApprovalResponse,
OpenAIResponseObject,
OpenAIResponseObjectStream,
OpenAIResponseOutput,
OpenAIResponseOutputMessageMCPListTools,
OpenAIResponseTool,
OpenAIResponseToolMCP,
)
from llama_stack.apis.inference import OpenAIChatCompletionToolCall, OpenAIMessageParam, OpenAIResponseFormatParam
@ -55,6 +63,86 @@ class ChatCompletionResult:
return bool(self.tool_calls)
class ToolContext(BaseModel):
"""Holds information about tools from this and (if relevant)
previous response in order to facilitate reuse of previous
listings where appropriate."""
# tools argument passed into current request:
current_tools: list[OpenAIResponseInputTool]
# reconstructed map of tool -> mcp server from previous response:
previous_tools: dict[str, OpenAIResponseInputToolMCP]
# reusable mcp-list-tools objects from previous response:
previous_tool_listings: list[OpenAIResponseOutputMessageMCPListTools]
# tool arguments from current request that still need to be processed:
tools_to_process: list[OpenAIResponseInputTool]
def __init__(
self,
current_tools: list[OpenAIResponseInputTool] | None,
):
super().__init__(
current_tools=current_tools or [],
previous_tools={},
previous_tool_listings=[],
tools_to_process=current_tools or [],
)
def recover_tools_from_previous_response(
self,
previous_response: OpenAIResponseObject,
):
"""Determine which mcp_list_tools objects from previous response we can reuse."""
if self.current_tools and previous_response.tools:
previous_tools_by_label: dict[str, OpenAIResponseToolMCP] = {}
for tool in previous_response.tools:
if isinstance(tool, OpenAIResponseToolMCP):
previous_tools_by_label[tool.server_label] = tool
# collect tool definitions which are the same in current and previous requests:
tools_to_process = []
matched: dict[str, OpenAIResponseInputToolMCP] = {}
for tool in self.current_tools:
if isinstance(tool, OpenAIResponseInputToolMCP) and tool.server_label in previous_tools_by_label:
previous_tool = previous_tools_by_label[tool.server_label]
if previous_tool.allowed_tools == tool.allowed_tools:
matched[tool.server_label] = tool
else:
tools_to_process.append(tool)
else:
tools_to_process.append(tool)
# tools that are not the same or were not previously defined need to be processed:
self.tools_to_process = tools_to_process
# for all matched definitions, get the mcp_list_tools objects from the previous output:
self.previous_tool_listings = [
obj for obj in previous_response.output if obj.type == "mcp_list_tools" and obj.server_label in matched
]
# reconstruct the tool to server mappings that can be reused:
for listing in self.previous_tool_listings:
definition = matched[listing.server_label]
for tool in listing.tools:
self.previous_tools[tool.name] = definition
def available_tools(self) -> list[OpenAIResponseTool]:
if not self.current_tools:
return []
def convert_tool(tool: OpenAIResponseInputTool) -> OpenAIResponseTool:
if isinstance(tool, OpenAIResponseInputToolWebSearch):
return tool
if isinstance(tool, OpenAIResponseInputToolFileSearch):
return tool
if isinstance(tool, OpenAIResponseInputToolFunction):
return tool
if isinstance(tool, OpenAIResponseInputToolMCP):
return OpenAIResponseToolMCP(
server_label=tool.server_label,
allowed_tools=tool.allowed_tools,
)
return [convert_tool(tool) for tool in self.current_tools]
class ChatCompletionContext(BaseModel):
model: str
messages: list[OpenAIMessageParam]
@ -62,6 +150,7 @@ class ChatCompletionContext(BaseModel):
chat_tools: list[ChatCompletionToolParam] | None = None
temperature: float | None
response_format: OpenAIResponseFormatParam
tool_context: ToolContext | None
approval_requests: list[OpenAIResponseMCPApprovalRequest] = []
approval_responses: dict[str, OpenAIResponseMCPApprovalResponse] = {}
@ -72,6 +161,7 @@ class ChatCompletionContext(BaseModel):
response_tools: list[OpenAIResponseInputTool] | None,
temperature: float | None,
response_format: OpenAIResponseFormatParam,
tool_context: ToolContext,
inputs: list[OpenAIResponseInput] | str,
):
super().__init__(
@ -80,6 +170,7 @@ class ChatCompletionContext(BaseModel):
response_tools=response_tools,
temperature=temperature,
response_format=response_format,
tool_context=tool_context,
)
if not isinstance(inputs, str):
self.approval_requests = [input for input in inputs if input.type == "mcp_approval_request"]
@ -96,3 +187,8 @@ class ChatCompletionContext(BaseModel):
if request.name == tool_name and request.arguments == arguments:
return request
return None
def available_tools(self) -> list[OpenAIResponseTool]:
if not self.tool_context:
return []
return self.tool_context.available_tools()

56
tests/unit/providers/agents/meta_reference/test_openai_responses.py
View file

@ -4,7 +4,7 @@
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
from unittest.mock import AsyncMock
from unittest.mock import AsyncMock, patch
import pytest
from openai.types.chat.chat_completion_chunk import (
@ -20,6 +20,7 @@ from llama_stack.apis.agents.openai_responses import (
ListOpenAIResponseInputItem,
OpenAIResponseInputMessageContentText,
OpenAIResponseInputToolFunction,
OpenAIResponseInputToolMCP,
OpenAIResponseInputToolWebSearch,
OpenAIResponseMessage,
OpenAIResponseOutputMessageContentOutputText,
@ -38,7 +39,7 @@ from llama_stack.apis.inference import (
OpenAIResponseFormatJSONSchema,
OpenAIUserMessageParam,
)
from llama_stack.apis.tools.tools import ToolDef, ToolGroups, ToolInvocationResult, ToolRuntime
from llama_stack.apis.tools.tools import ListToolDefsResponse, ToolDef, ToolGroups, ToolInvocationResult, ToolRuntime
from llama_stack.core.access_control.access_control import default_policy
from llama_stack.core.datatypes import ResponsesStoreConfig
from llama_stack.providers.inline.agents.meta_reference.responses.openai_responses import (
@ -963,6 +964,57 @@ async def test_store_response_uses_rehydrated_input_with_previous_response(
assert result.status == "completed"
@patch("llama_stack.providers.utils.tools.mcp.list_mcp_tools")
async def test_reuse_mcp_tool_list(
mock_list_mcp_tools, openai_responses_impl, mock_responses_store, mock_inference_api
):
"""Test that mcp_list_tools can be reused where appropriate."""
mock_inference_api.openai_chat_completion.return_value = fake_stream()
mock_list_mcp_tools.return_value = ListToolDefsResponse(
data=[ToolDef(name="test_tool", description="a test tool", input_schema={}, output_schema={})]
)
res1 = await openai_responses_impl.create_openai_response(
input="What is 2+2?",
model="meta-llama/Llama-3.1-8B-Instruct",
store=True,
tools=[
OpenAIResponseInputToolFunction(name="fake", parameters=None),
OpenAIResponseInputToolMCP(server_label="alabel", server_url="aurl"),
],
)
args = mock_responses_store.store_response_object.call_args
data = args.kwargs["response_object"].model_dump()
data["input"] = [input_item.model_dump() for input_item in args.kwargs["input"]]
data["messages"] = [msg.model_dump() for msg in args.kwargs["messages"]]
stored = _OpenAIResponseObjectWithInputAndMessages(**data)
mock_responses_store.get_response_object.return_value = stored
res2 = await openai_responses_impl.create_openai_response(
previous_response_id=res1.id,
input="Now what is 3+3?",
model="meta-llama/Llama-3.1-8B-Instruct",
store=True,
tools=[
OpenAIResponseInputToolMCP(server_label="alabel", server_url="aurl"),
],
)
assert len(mock_inference_api.openai_chat_completion.call_args_list) == 2
second_call = mock_inference_api.openai_chat_completion.call_args_list[1]
tools_seen = second_call.kwargs["tools"]
assert len(tools_seen) == 1
assert tools_seen[0]["function"]["name"] == "test_tool"
assert tools_seen[0]["function"]["description"] == "a test tool"
assert mock_list_mcp_tools.call_count == 1
listings = [obj for obj in res2.output if obj.type == "mcp_list_tools"]
assert len(listings) == 1
assert listings[0].server_label == "alabel"
assert len(listings[0].tools) == 1
assert listings[0].tools[0].name == "test_tool"
@pytest.mark.parametrize(
"text_format, response_format",
[

183
tests/unit/providers/agents/meta_reference/test_response_tool_context.py Normal file
View file

@ -0,0 +1,183 @@
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
from llama_stack.apis.agents.openai_responses import (
MCPListToolsTool,
OpenAIResponseInputToolFileSearch,
OpenAIResponseInputToolFunction,
OpenAIResponseInputToolMCP,
OpenAIResponseInputToolWebSearch,
OpenAIResponseObject,
OpenAIResponseOutputMessageMCPListTools,
OpenAIResponseToolMCP,
)
from llama_stack.providers.inline.agents.meta_reference.responses.types import ToolContext
class TestToolContext:
def test_no_tools(self):
tools = []
context = ToolContext(tools)
previous_response = OpenAIResponseObject(created_at=1234, id="test", model="mymodel", output=[], status="")
context.recover_tools_from_previous_response(previous_response)
assert len(context.tools_to_process) == 0
assert len(context.previous_tools) == 0
assert len(context.previous_tool_listings) == 0
def test_no_previous_tools(self):
tools = [
OpenAIResponseInputToolFileSearch(vector_store_ids=["fake"]),
OpenAIResponseInputToolMCP(server_label="label", server_url="url"),
]
context = ToolContext(tools)
previous_response = OpenAIResponseObject(created_at=1234, id="test", model="mymodel", output=[], status="")
context.recover_tools_from_previous_response(previous_response)
assert len(context.tools_to_process) == 2
assert len(context.previous_tools) == 0
assert len(context.previous_tool_listings) == 0
def test_reusable_server(self):
tools = [
OpenAIResponseInputToolFileSearch(vector_store_ids=["fake"]),
OpenAIResponseInputToolMCP(server_label="alabel", server_url="aurl"),
]
context = ToolContext(tools)
output = [
OpenAIResponseOutputMessageMCPListTools(
id="test", server_label="alabel", tools=[MCPListToolsTool(name="test_tool", input_schema={})]
)
]
previous_response = OpenAIResponseObject(created_at=1234, id="test", model="fake", output=output, status="")
previous_response.tools = [
OpenAIResponseInputToolFileSearch(vector_store_ids=["fake"]),
OpenAIResponseToolMCP(server_label="alabel"),
]
context.recover_tools_from_previous_response(previous_response)
assert len(context.tools_to_process) == 1
assert context.tools_to_process[0].type == "file_search"
assert len(context.previous_tools) == 1
assert context.previous_tools["test_tool"].server_label == "alabel"
assert context.previous_tools["test_tool"].server_url == "aurl"
assert len(context.previous_tool_listings) == 1
assert len(context.previous_tool_listings[0].tools) == 1
assert context.previous_tool_listings[0].server_label == "alabel"
def test_multiple_reusable_servers(self):
tools = [
OpenAIResponseInputToolFunction(name="fake", parameters=None),
OpenAIResponseInputToolMCP(server_label="anotherlabel", server_url="anotherurl"),
OpenAIResponseInputToolWebSearch(),
OpenAIResponseInputToolMCP(server_label="alabel", server_url="aurl"),
]
context = ToolContext(tools)
output = [
OpenAIResponseOutputMessageMCPListTools(
id="test1", server_label="alabel", tools=[MCPListToolsTool(name="test_tool", input_schema={})]
),
OpenAIResponseOutputMessageMCPListTools(
id="test2",
server_label="anotherlabel",
tools=[MCPListToolsTool(name="some_other_tool", input_schema={})],
),
]
previous_response = OpenAIResponseObject(created_at=1234, id="test", model="fake", output=output, status="")
previous_response.tools = [
OpenAIResponseInputToolFunction(name="fake", parameters=None),
OpenAIResponseToolMCP(server_label="anotherlabel", server_url="anotherurl"),
OpenAIResponseInputToolWebSearch(type="web_search"),
OpenAIResponseToolMCP(server_label="alabel", server_url="aurl"),
]
context.recover_tools_from_previous_response(previous_response)
assert len(context.tools_to_process) == 2
assert context.tools_to_process[0].type == "function"
assert context.tools_to_process[1].type == "web_search"
assert len(context.previous_tools) == 2
assert context.previous_tools["test_tool"].server_label == "alabel"
assert context.previous_tools["test_tool"].server_url == "aurl"
assert context.previous_tools["some_other_tool"].server_label == "anotherlabel"
assert context.previous_tools["some_other_tool"].server_url == "anotherurl"
assert len(context.previous_tool_listings) == 2
assert len(context.previous_tool_listings[0].tools) == 1
assert context.previous_tool_listings[0].server_label == "alabel"
assert len(context.previous_tool_listings[1].tools) == 1
assert context.previous_tool_listings[1].server_label == "anotherlabel"
def test_multiple_servers_only_one_reusable(self):
tools = [
OpenAIResponseInputToolFunction(name="fake", parameters=None),
OpenAIResponseInputToolMCP(server_label="anotherlabel", server_url="anotherurl"),
OpenAIResponseInputToolWebSearch(type="web_search"),
OpenAIResponseInputToolMCP(server_label="alabel", server_url="aurl"),
]
context = ToolContext(tools)
output = [
OpenAIResponseOutputMessageMCPListTools(
id="test2",
server_label="anotherlabel",
tools=[MCPListToolsTool(name="some_other_tool", input_schema={})],
)
]
previous_response = OpenAIResponseObject(created_at=1234, id="test", model="fake", output=output, status="")
previous_response.tools = [
OpenAIResponseInputToolFunction(name="fake", parameters=None),
OpenAIResponseToolMCP(server_label="anotherlabel", server_url="anotherurl"),
OpenAIResponseInputToolWebSearch(type="web_search"),
]
context.recover_tools_from_previous_response(previous_response)
assert len(context.tools_to_process) == 3
assert context.tools_to_process[0].type == "function"
assert context.tools_to_process[1].type == "web_search"
assert context.tools_to_process[2].type == "mcp"
assert len(context.previous_tools) == 1
assert context.previous_tools["some_other_tool"].server_label == "anotherlabel"
assert context.previous_tools["some_other_tool"].server_url == "anotherurl"
assert len(context.previous_tool_listings) == 1
assert len(context.previous_tool_listings[0].tools) == 1
assert context.previous_tool_listings[0].server_label == "anotherlabel"
def test_mismatched_allowed_tools(self):
tools = [
OpenAIResponseInputToolFunction(name="fake", parameters=None),
OpenAIResponseInputToolMCP(server_label="anotherlabel", server_url="anotherurl"),
OpenAIResponseInputToolWebSearch(type="web_search"),
OpenAIResponseInputToolMCP(server_label="alabel", server_url="aurl", allowed_tools=["test_tool_2"]),
]
context = ToolContext(tools)
output = [
OpenAIResponseOutputMessageMCPListTools(
id="test1", server_label="alabel", tools=[MCPListToolsTool(name="test_tool_1", input_schema={})]
),
OpenAIResponseOutputMessageMCPListTools(
id="test2",
server_label="anotherlabel",
tools=[MCPListToolsTool(name="some_other_tool", input_schema={})],
),
]
previous_response = OpenAIResponseObject(created_at=1234, id="test", model="fake", output=output, status="")
previous_response.tools = [
OpenAIResponseInputToolFunction(name="fake", parameters=None),
OpenAIResponseToolMCP(server_label="anotherlabel", server_url="anotherurl"),
OpenAIResponseInputToolWebSearch(type="web_search"),
OpenAIResponseToolMCP(server_label="alabel", server_url="aurl"),
]
context.recover_tools_from_previous_response(previous_response)
assert len(context.tools_to_process) == 3
assert context.tools_to_process[0].type == "function"
assert context.tools_to_process[1].type == "web_search"
assert context.tools_to_process[2].type == "mcp"
assert len(context.previous_tools) == 1
assert context.previous_tools["some_other_tool"].server_label == "anotherlabel"
assert context.previous_tools["some_other_tool"].server_url == "anotherurl"
assert len(context.previous_tool_listings) == 1
assert len(context.previous_tool_listings[0].tools) == 1
assert context.previous_tool_listings[0].server_label == "anotherlabel"