Skip to content

armgit / letta

Public Imported from GitHub
0

Platform for stateful agents: AI with advanced memory that can learn and self-improve over time.

feat: revert model_settings (#6089)

Sarah Woodersddc87418f43c
diff --git a/fern/openapi.json b/fern/openapi.json
index ad30f2474..617e19bf0 100644
--- a/fern/openapi.json
+++ b/fern/openapi.json
@@ -18937,84 +18937,24 @@
"model": {
"anyOf": [
{
- "type": "string"
+ "$ref": "#/components/schemas/ModelSettings"
},
{
"type": "null"
}
],
- "title": "Model",
- "description": "The model handle used by the agent (format: provider/model-name)."
+ "description": "The model used by the agent."
},
"embedding": {
"anyOf": [
{
- "type": "string"
+ "$ref": "#/components/schemas/EmbeddingModelSettings"
},
{
"type": "null"
}
],
- "title": "Embedding",
- "description": "The embedding model handle used by the agent (format: provider/model-name)."
- },
- "model_settings": {
- "anyOf": [
- {
- "oneOf": [
- {
- "$ref": "#/components/schemas/OpenAIModelSettings"
- },
- {
- "$ref": "#/components/schemas/AnthropicModelSettings"
- },
- {
- "$ref": "#/components/schemas/GoogleAIModelSettings"
- },
- {
- "$ref": "#/components/schemas/GoogleVertexModelSettings"
- },
- {
- "$ref": "#/components/schemas/AzureModelSettings"
- },
- {
- "$ref": "#/components/schemas/XAIModelSettings"
- },
- {
- "$ref": "#/components/schemas/GroqModelSettings"
- },
- {
- "$ref": "#/components/schemas/DeepseekModelSettings"
- },
- {
- "$ref": "#/components/schemas/TogetherModelSettings"
- },
- {
- "$ref": "#/components/schemas/BedrockModelSettings"
- }
- ],
- "discriminator": {
- "propertyName": "provider",
- "mapping": {
- "anthropic": "#/components/schemas/AnthropicModelSettings",
- "azure": "#/components/schemas/AzureModelSettings",
- "bedrock": "#/components/schemas/BedrockModelSettings",
- "deepseek": "#/components/schemas/DeepseekModelSettings",
- "google_ai": "#/components/schemas/GoogleAIModelSettings",
- "google_vertex": "#/components/schemas/GoogleVertexModelSettings",
- "groq": "#/components/schemas/GroqModelSettings",
- "openai": "#/components/schemas/OpenAIModelSettings",
- "together": "#/components/schemas/TogetherModelSettings",
- "xai": "#/components/schemas/XAIModelSettings"
- }
- }
- },
- {
- "type": "null"
- }
- ],
- "title": "Model Settings",
- "description": "The model settings used by the agent."
+ "description": "The embedding model used by the agent."
},
"response_format": {
"anyOf": [
@@ -19406,6 +19346,11 @@
},
"AnthropicModelSettings": {
"properties": {
+ "model": {
+ "type": "string",
+ "title": "Model",
+ "description": "The name of the model."
+ },
"max_output_tokens": {
"type": "integer",
"title": "Max Output Tokens",
@@ -19454,6 +19399,7 @@
}
},
"type": "object",
+ "required": ["model"],
"title": "AnthropicModelSettings"
},
"AnthropicThinking": {
@@ -19467,8 +19413,6 @@
},
"budget_tokens": {
"type": "integer",
- "maximum": 1024,
- "minimum": 0,
"title": "Budget Tokens",
"description": "The maximum number of tokens the model can use for extended thinking.",
"default": 1024
@@ -20277,6 +20221,11 @@
},
"AzureModelSettings": {
"properties": {
+ "model": {
+ "type": "string",
+ "title": "Model",
+ "description": "The name of the model."
+ },
"max_output_tokens": {
"type": "integer",
"title": "Max Output Tokens",
@@ -20298,8 +20247,6 @@
},
"temperature": {
"type": "number",
- "maximum": 1,
- "minimum": 0,
"title": "Temperature",
"description": "The temperature of the model.",
"default": 0.7
@@ -20336,6 +20283,7 @@
}
},
"type": "object",
+ "required": ["model"],
"title": "AzureModelSettings",
"description": "Azure OpenAI model configuration (OpenAI-compatible)."
},
@@ -20592,6 +20540,11 @@
},
"BedrockModelSettings": {
"properties": {
+ "model": {
+ "type": "string",
+ "title": "Model",
+ "description": "The name of the model."
+ },
"max_output_tokens": {
"type": "integer",
"title": "Max Output Tokens",
@@ -20613,8 +20566,6 @@
},
"temperature": {
"type": "number",
- "maximum": 1,
- "minimum": 0,
"title": "Temperature",
"description": "The temperature of the model.",
"default": 0.7
@@ -20651,6 +20602,7 @@
}
},
"type": "object",
+ "required": ["model"],
"title": "BedrockModelSettings",
"description": "AWS Bedrock model configuration."
},
@@ -23207,27 +23159,6 @@
{
"type": "string"
},
- {
- "type": "null"
- }
- ],
- "title": "Model",
- "description": "The model handle for the agent to use (format: provider/model-name)."
- },
- "embedding": {
- "anyOf": [
- {
- "type": "string"
- },
- {
- "type": "null"
- }
- ],
- "title": "Embedding",
- "description": "The embedding model handle used by the agent (format: provider/model-name)."
- },
- "model_settings": {
- "anyOf": [
{
"oneOf": [
{
@@ -23281,8 +23212,23 @@
"type": "null"
}
],
- "title": "Model Settings",
- "description": "The model settings for the agent."
+ "title": "Model",
+ "description": "The model handle or model settings for the agent to use, specified either by a handle or an object. See the model schema for more information."
+ },
+ "embedding": {
+ "anyOf": [
+ {
+ "type": "string"
+ },
+ {
+ "$ref": "#/components/schemas/EmbeddingModelSettings"
+ },
+ {
+ "type": "null"
+ }
+ ],
+ "title": "Embedding",
+ "description": "The embedding configuration handle used by the agent, specified in the format provider/model-name."
},
"context_window_limit": {
"anyOf": [
@@ -24041,6 +23987,11 @@
},
"DeepseekModelSettings": {
"properties": {
+ "model": {
+ "type": "string",
+ "title": "Model",
+ "description": "The name of the model."
+ },
"max_output_tokens": {
"type": "integer",
"title": "Max Output Tokens",
@@ -24062,8 +24013,6 @@
},
"temperature": {
"type": "number",
- "maximum": 1,
- "minimum": 0,
"title": "Temperature",
"description": "The temperature of the model.",
"default": 0.7
@@ -24100,6 +24049,7 @@
}
},
"type": "object",
+ "required": ["model"],
"title": "DeepseekModelSettings",
"description": "Deepseek model configuration (OpenAI-compatible)."
},
@@ -24628,6 +24578,25 @@
],
"title": "EmbeddingModel"
},
+ "EmbeddingModelSettings": {
+ "properties": {
+ "model": {
+ "type": "string",
+ "title": "Model",
+ "description": "The name of the model."
+ },
+ "provider": {
+ "type": "string",
+ "enum": ["openai", "ollama"],
+ "title": "Provider",
+ "description": "The provider of the model."
+ }
+ },
+ "type": "object",
+ "required": ["model", "provider"],
+ "title": "EmbeddingModelSettings",
+ "description": "Schema for defining settings for an embedding model"
+ },
"EventMessage": {
"properties": {
"id": {
@@ -25696,8 +25665,6 @@
},
"thinking_budget": {
"type": "integer",
- "maximum": 1024,
- "minimum": 0,
"title": "Thinking Budget",
"description": "The thinking budget for the model.",
"default": 1024
@@ -25779,6 +25746,11 @@
},
"GoogleAIModelSettings": {
"properties": {
+ "model": {
+ "type": "string",
+ "title": "Model",
+ "description": "The name of the model."
+ },
"max_output_tokens": {
"type": "integer",
"title": "Max Output Tokens",
@@ -25800,8 +25772,6 @@
},
"temperature": {
"type": "number",
- "maximum": 1,
- "minimum": 0,
"title": "Temperature",
"description": "The temperature of the model.",
"default": 0.7
@@ -25846,10 +25816,16 @@
}
},
"type": "object",
+ "required": ["model"],
"title": "GoogleAIModelSettings"
},
"GoogleVertexModelSettings": {
"properties": {
+ "model": {
+ "type": "string",
+ "title": "Model",
+ "description": "The name of the model."
+ },
"max_output_tokens": {
"type": "integer",
"title": "Max Output Tokens",
@@ -25871,8 +25847,6 @@
},
"temperature": {
"type": "number",
- "maximum": 1,
- "minimum": 0,
"title": "Temperature",
"description": "The temperature of the model.",
"default": 0.7
@@ -25917,10 +25891,16 @@
}
},
"type": "object",
+ "required": ["model"],
"title": "GoogleVertexModelSettings"
},
"GroqModelSettings": {
"properties": {
+ "model": {
+ "type": "string",
+ "title": "Model",
+ "description": "The name of the model."
+ },
"max_output_tokens": {
"type": "integer",
"title": "Max Output Tokens",
@@ -25942,8 +25922,6 @@
},
"temperature": {
"type": "number",
- "maximum": 1,
- "minimum": 0,
"title": "Temperature",
"description": "The temperature of the model.",
"default": 0.7
@@ -25980,6 +25958,7 @@
}
},
"type": "object",
+ "required": ["model"],
"title": "GroqModelSettings",
"description": "Groq model configuration (OpenAI-compatible)."
},
@@ -27382,27 +27361,6 @@
{
"type": "string"
},
- {
- "type": "null"
- }
- ],
- "title": "Model",
- "description": "The model handle for the agent to use (format: provider/model-name)."
- },
- "embedding": {
- "anyOf": [
- {
- "type": "string"
- },
- {
- "type": "null"
- }
- ],
- "title": "Embedding",
- "description": "The embedding model handle used by the agent (format: provider/model-name)."
- },
- "model_settings": {
- "anyOf": [
{
"oneOf": [
{
@@ -27456,8 +27414,23 @@
"type": "null"
}
],
- "title": "Model Settings",
- "description": "The model settings for the agent."
+ "title": "Model",
+ "description": "The model handle or model settings for the agent to use, specified either by a handle or an object. See the model schema for more information."
+ },
+ "embedding": {
+ "anyOf": [
+ {
+ "type": "string"
+ },
+ {
+ "$ref": "#/components/schemas/EmbeddingModelSettings"
+ },
+ {
+ "type": "null"
+ }
+ ],
+ "title": "Embedding",
+ "description": "The embedding configuration handle used by the agent, specified in the format provider/model-name."
},
"context_window_limit": {
"anyOf": [
@@ -30655,6 +30628,31 @@
],
"title": "Model"
},
+ "ModelSettings": {
+ "properties": {
+ "model": {
+ "type": "string",
+ "title": "Model",
+ "description": "The name of the model."
+ },
+ "max_output_tokens": {
+ "type": "integer",
+ "title": "Max Output Tokens",
+ "description": "The maximum number of tokens the model can generate.",
+ "default": 4096
+ },
+ "parallel_tool_calls": {
+ "type": "boolean",
+ "title": "Parallel Tool Calls",
+ "description": "Whether to enable parallel tool calling.",
+ "default": false
+ }
+ },
+ "type": "object",
+ "required": ["model"],
+ "title": "ModelSettings",
+ "description": "Schema for defining settings for a model"
+ },
"ModifyApprovalRequest": {
"properties": {
"requires_approval": {
@@ -30754,6 +30752,11 @@
},
"OpenAIModelSettings": {
"properties": {
+ "model": {
+ "type": "string",
+ "title": "Model",
+ "description": "The name of the model."
+ },
"max_output_tokens": {
"type": "integer",
"title": "Max Output Tokens",
@@ -30775,8 +30778,6 @@
},
"temperature": {
"type": "number",
- "maximum": 1,
- "minimum": 0,
"title": "Temperature",
"description": "The temperature of the model.",
"default": 0.7
@@ -30820,6 +30821,7 @@
}
},
"type": "object",
+ "required": ["model"],
"title": "OpenAIModelSettings"
},
"OpenAIReasoning": {
@@ -34395,6 +34397,11 @@
},
"TogetherModelSettings": {
"properties": {
+ "model": {
+ "type": "string",
+ "title": "Model",
+ "description": "The name of the model."
+ },
"max_output_tokens": {
"type": "integer",
"title": "Max Output Tokens",
@@ -34416,8 +34423,6 @@
},
"temperature": {
"type": "number",
- "maximum": 1,
- "minimum": 0,
"title": "Temperature",
"description": "The temperature of the model.",
"default": 0.7
@@ -34454,6 +34459,7 @@
}
},
"type": "object",
+ "required": ["model"],
"title": "TogetherModelSettings",
"description": "Together AI model configuration (OpenAI-compatible)."
},
@@ -36057,27 +36063,6 @@
{
"type": "string"
},
- {
- "type": "null"
- }
- ],
- "title": "Model",
- "description": "The model handle used by the agent (format: provider/model-name)."
- },
- "embedding": {
- "anyOf": [
- {
- "type": "string"
- },
- {
- "type": "null"
- }
- ],
- "title": "Embedding",
- "description": "The embedding model handle used by the agent (format: provider/model-name)."
- },
- "model_settings": {
- "anyOf": [
{
"oneOf": [
{
@@ -36131,8 +36116,23 @@
"type": "null"
}
],
- "title": "Model Settings",
- "description": "The model settings for the agent."
+ "title": "Model",
+ "description": "The model used by the agent, specified either by a handle or an object. See the model schema for more information."
+ },
+ "embedding": {
+ "anyOf": [
+ {
+ "type": "string"
+ },
+ {
+ "$ref": "#/components/schemas/EmbeddingModelSettings"
+ },
+ {
+ "type": "null"
+ }
+ ],
+ "title": "Embedding",
+ "description": "The embedding configuration handle used by the agent, specified in the format provider/model-name."
},
"context_window_limit": {
"anyOf": [
@@ -36849,6 +36849,11 @@
},
"XAIModelSettings": {
"properties": {
+ "model": {
+ "type": "string",
+ "title": "Model",
+ "description": "The name of the model."
+ },
"max_output_tokens": {
"type": "integer",
"title": "Max Output Tokens",
@@ -36870,8 +36875,6 @@
},
"temperature": {
"type": "number",
- "maximum": 1,
- "minimum": 0,
"title": "Temperature",
"description": "The temperature of the model.",
"default": 0.7
@@ -36908,6 +36911,7 @@
}
},
"type": "object",
+ "required": ["model"],
"title": "XAIModelSettings",
"description": "xAI model configuration (OpenAI-compatible)."
},
@@ -37181,27 +37185,6 @@
{
"type": "string"
},
- {
- "type": "null"
- }
- ],
- "title": "Model",
- "description": "The model handle for the agent to use (format: provider/model-name)."
- },
- "embedding": {
- "anyOf": [
- {
- "type": "string"
- },
- {
- "type": "null"
- }
- ],
- "title": "Embedding",
- "description": "The embedding model handle used by the agent (format: provider/model-name)."
- },
- "model_settings": {
- "anyOf": [
{
"oneOf": [
{
@@ -37255,8 +37238,23 @@
"type": "null"
}
],
- "title": "Model Settings",
- "description": "The model settings for the agent."
+ "title": "Model",
+ "description": "The model handle or model settings for the agent to use, specified either by a handle or an object. See the model schema for more information."
+ },
+ "embedding": {
+ "anyOf": [
+ {
+ "type": "string"
+ },
+ {
+ "$ref": "#/components/schemas/EmbeddingModelSettings"
+ },
+ {
+ "type": "null"
+ }
+ ],
+ "title": "Embedding",
+ "description": "The embedding configuration handle used by the agent, specified in the format provider/model-name."
},
"context_window_limit": {
"anyOf": [
diff --git a/letta/orm/agent.py b/letta/orm/agent.py
index 01b3cbcaa..22fd33de6 100644
--- a/letta/orm/agent.py
+++ b/letta/orm/agent.py
@@ -285,9 +285,7 @@ class Agent(SqlalchemyBase, OrganizationMixin, ProjectMixin, TemplateEntityMixin
if resolver:
state[field_name] = resolver()
- state["model"] = self.llm_config.handle if self.llm_config else None
- state["model_settings"] = self.llm_config._to_model_settings() if self.llm_config else None
- state["embedding"] = self.embedding_config.handle if self.embedding_config else None
+ state["model"] = self.llm_config._to_model() if self.llm_config else None
return self.__pydantic_model__(**state)
@@ -427,8 +425,6 @@ class Agent(SqlalchemyBase, OrganizationMixin, ProjectMixin, TemplateEntityMixin
state["managed_group"] = multi_agent_group
state["tool_exec_environment_variables"] = tool_exec_environment_variables
state["secrets"] = tool_exec_environment_variables
- state["model"] = self.llm_config.handle if self.llm_config else None
- state["model_settings"] = self.llm_config._to_model_settings() if self.llm_config else None
- state["embedding"] = self.embedding_config.handle if self.embedding_config else None
+ state["model"] = self.llm_config._to_model() if self.llm_config else None
return self.__pydantic_model__(**state)
diff --git a/letta/schemas/agent.py b/letta/schemas/agent.py
index b3fa1cab1..1f1aa7b3e 100644
--- a/letta/schemas/agent.py
+++ b/letta/schemas/agent.py
@@ -5,7 +5,7 @@ from typing import Dict, List, Literal, Optional
from pydantic import BaseModel, ConfigDict, Field, field_validator, model_validator
from letta.constants import CORE_MEMORY_LINE_NUMBER_WARNING, DEFAULT_EMBEDDING_CHUNK_SIZE
-from letta.errors import AgentExportProcessingError, LettaInvalidArgumentError
+from letta.errors import AgentExportProcessingError
from letta.schemas.block import Block, CreateBlock
from letta.schemas.embedding_config import EmbeddingConfig
from letta.schemas.enums import PrimitiveType
@@ -18,7 +18,7 @@ from letta.schemas.letta_stop_reason import StopReasonType
from letta.schemas.llm_config import LLMConfig
from letta.schemas.memory import Memory
from letta.schemas.message import Message, MessageCreate
-from letta.schemas.model import ModelSettingsUnion
+from letta.schemas.model import EmbeddingModelSettings, ModelSettings, ModelSettingsUnion
from letta.schemas.openai.chat_completion_response import UsageStatistics
from letta.schemas.response_format import ResponseFormatUnion
from letta.schemas.source import Source
@@ -83,9 +83,8 @@ class AgentState(OrmMetadataBase, validate_assignment=True):
embedding_config: EmbeddingConfig = Field(
..., description="Deprecated: Use `embedding` field instead. The embedding configuration used by the agent.", deprecated=True
)
- model: Optional[str] = Field(None, description="The model handle used by the agent (format: provider/model-name).")
- embedding: Optional[str] = Field(None, description="The embedding model handle used by the agent (format: provider/model-name).")
- model_settings: Optional[ModelSettingsUnion] = Field(None, description="The model settings used by the agent.")
+ model: Optional[ModelSettings] = Field(None, description="The model used by the agent.")
+ embedding: Optional[EmbeddingModelSettings] = Field(None, description="The embedding model used by the agent.")
response_format: Optional[ResponseFormatUnion] = Field(
None,
@@ -230,12 +229,13 @@ class CreateAgent(BaseModel, validate_assignment=True): #
embedding_config: Optional[EmbeddingConfig] = Field(
None, description="Deprecated: Use `embedding` field instead. The embedding configuration used by the agent.", deprecated=True
)
- model: Optional[str] = Field( # TODO: make this required (breaking change)
+ model: Optional[str | ModelSettingsUnion] = Field( # TODO: make this required (breaking change)
None,
- description="The model handle for the agent to use (format: provider/model-name).",
+ description="The model handle or model settings for the agent to use, specified either by a handle or an object. See the model schema for more information.",
+ )
+ embedding: Optional[str | EmbeddingModelSettings] = Field(
+ None, description="The embedding configuration handle used by the agent, specified in the format provider/model-name."
)
- embedding: Optional[str] = Field(None, description="The embedding model handle used by the agent (format: provider/model-name).")
- model_settings: Optional[ModelSettingsUnion] = Field(None, description="The model settings for the agent.")
context_window_limit: Optional[int] = Field(None, description="The context window limit used by the agent.")
embedding_chunk_size: Optional[int] = Field(
@@ -348,12 +348,9 @@ class CreateAgent(BaseModel, validate_assignment=True): #
if not model:
return model
- if "/" not in model:
- raise LettaInvalidArgumentError("The model handle should be in the format provider/model-name", argument_name="model")
-
provider_name, model_name = model.split("/", 1)
if not provider_name or not model_name:
- raise LettaInvalidArgumentError("The model handle should be in the format provider/model-name", argument_name="model")
+ raise ValueError("The llm config handle should be in the format provider/model-name")
return model
@@ -363,12 +360,9 @@ class CreateAgent(BaseModel, validate_assignment=True): #
if not embedding:
return embedding
- if "/" not in embedding:
- raise ValueError("The embedding handle should be in the format provider/model-name")
-
provider_name, embedding_name = embedding.split("/", 1)
if not provider_name or not embedding_name:
- raise ValueError("The embedding handle should be in the format provider/model-name")
+ raise ValueError("The embedding config handle should be in the format provider/model-name")
return embedding
@@ -416,12 +410,13 @@ class UpdateAgent(BaseModel):
)
# model configuration
- model: Optional[str] = Field(
+ model: Optional[str | ModelSettingsUnion ] = Field(
None,
- description="The model handle used by the agent (format: provider/model-name).",
+ description="The model used by the agent, specified either by a handle or an object. See the model schema for more information.",
+ )
+ embedding: Optional[str | EmbeddingModelSettings] = Field(
+ None, description="The embedding configuration handle used by the agent, specified in the format provider/model-name."
)
- embedding: Optional[str] = Field(None, description="The embedding model handle used by the agent (format: provider/model-name).")
- model_settings: Optional[ModelSettingsUnion] = Field(None, description="The model settings for the agent.")
context_window_limit: Optional[int] = Field(None, description="The context window limit used by the agent.")
reasoning: Optional[bool] = Field(
None,
diff --git a/letta/schemas/llm_config.py b/letta/schemas/llm_config.py
index 3c1c95fe5..d586a0df6 100644
--- a/letta/schemas/llm_config.py
+++ b/letta/schemas/llm_config.py
@@ -255,7 +255,7 @@ class LLMConfig(BaseModel):
+ (f" [ip={self.model_endpoint}]" if self.model_endpoint else "")
)
- def _to_model_settings(self) -> "ModelSettings":
+ def _to_model(self) -> "ModelSettings":
"""
Convert LLMConfig back into a Model schema (OpenAIModelSettings, AnthropicModelSettings, etc.).
This is the inverse of the _to_legacy_config_params() methods in model.py.
@@ -279,6 +279,7 @@ class LLMConfig(BaseModel):
if self.model_endpoint_type == "openai":
return OpenAIModelSettings(
+ model=self.model,
max_output_tokens=self.max_tokens or 4096,
temperature=self.temperature,
reasoning=OpenAIReasoning(reasoning_effort=self.reasoning_effort or "minimal"),
@@ -286,6 +287,7 @@ class LLMConfig(BaseModel):
elif self.model_endpoint_type == "anthropic":
thinking_type = "enabled" if self.enable_reasoner else "disabled"
return AnthropicModelSettings(
+ model=self.model,
max_output_tokens=self.max_tokens or 4096,
temperature=self.temperature,
thinking=AnthropicThinking(type=thinking_type, budget_tokens=self.max_reasoning_tokens or 1024),
@@ -293,6 +295,7 @@ class LLMConfig(BaseModel):
)
elif self.model_endpoint_type == "google_ai":
return GoogleAIModelSettings(
+ model=self.model,
max_output_tokens=self.max_tokens or 65536,
temperature=self.temperature,
thinking_config=GeminiThinkingConfig(
@@ -301,6 +304,7 @@ class LLMConfig(BaseModel):
)
elif self.model_endpoint_type == "google_vertex":
return GoogleVertexModelSettings(
+ model=self.model,
max_output_tokens=self.max_tokens or 65536,
temperature=self.temperature,
thinking_config=GeminiThinkingConfig(
@@ -309,34 +313,39 @@ class LLMConfig(BaseModel):
)
elif self.model_endpoint_type == "azure":
return AzureModelSettings(
+ model=self.model,
max_output_tokens=self.max_tokens or 4096,
temperature=self.temperature,
)
elif self.model_endpoint_type == "xai":
return XAIModelSettings(
+ model=self.model,
max_output_tokens=self.max_tokens or 4096,
temperature=self.temperature,
)
elif self.model_endpoint_type == "groq":
return GroqModelSettings(
+ model=self.model,
max_output_tokens=self.max_tokens or 4096,
temperature=self.temperature,
)
elif self.model_endpoint_type == "deepseek":
return DeepseekModelSettings(
+ model=self.model,
max_output_tokens=self.max_tokens or 4096,
temperature=self.temperature,
)
elif self.model_endpoint_type == "together":
return TogetherModelSettings(
+ model=self.model,
max_output_tokens=self.max_tokens or 4096,
temperature=self.temperature,
)
elif self.model_endpoint_type == "bedrock":
- return Model(max_output_tokens=self.max_tokens or 4096)
+ return Model(model=self.model, max_output_tokens=self.max_tokens or 4096)
else:
# If we don't know the model type, use the default Model schema
- return Model(max_output_tokens=self.max_tokens or 4096)
+ return Model(model=self.model, max_output_tokens=self.max_tokens or 4096)
@classmethod
def is_openai_reasoning_model(cls, config: "LLMConfig") -> bool:
diff --git a/letta/schemas/model.py b/letta/schemas/model.py
index 3c35af7d4..5bcb8e5fd 100644
--- a/letta/schemas/model.py
+++ b/letta/schemas/model.py
@@ -120,25 +120,6 @@ class Model(LLMConfig, ModelBase):
provider_category=llm_config.provider_category,
)
- @property
- def model_settings_schema(self) -> Optional[dict]:
- """Returns the JSON schema for the ModelSettings class corresponding to this model's provider."""
- PROVIDER_SETTINGS_MAP = {
- ProviderType.openai: OpenAIModelSettings,
- ProviderType.anthropic: AnthropicModelSettings,
- ProviderType.google_ai: GoogleAIModelSettings,
- ProviderType.google_vertex: GoogleVertexModelSettings,
- ProviderType.azure: AzureModelSettings,
- ProviderType.xai: XAIModelSettings,
- ProviderType.groq: GroqModelSettings,
- ProviderType.deepseek: DeepseekModelSettings,
- ProviderType.together: TogetherModelSettings,
- ProviderType.bedrock: BedrockModelSettings,
- }
-
- settings_class = PROVIDER_SETTINGS_MAP.get(self.provider_type)
- return settings_class.model_json_schema() if settings_class else None
-
class EmbeddingModel(EmbeddingConfig, ModelBase):
model_type: Literal["embedding"] = Field("embedding", description="Type of model (llm or embedding)")
@@ -203,11 +184,18 @@ class EmbeddingModel(EmbeddingConfig, ModelBase):
class ModelSettings(BaseModel):
"""Schema for defining settings for a model"""
- # model: str = Field(..., description="The name of the model.")
+ model: str = Field(..., description="The name of the model.")
max_output_tokens: int = Field(4096, description="The maximum number of tokens the model can generate.")
parallel_tool_calls: bool = Field(False, description="Whether to enable parallel tool calling.")
+class EmbeddingModelSettings(BaseModel):
+ """Schema for defining settings for an embedding model"""
+
+ model: str = Field(..., description="The name of the model.")
+ provider: Literal["openai", "ollama"] = Field(..., description="The provider of the model.")
+
+
class OpenAIReasoning(BaseModel):
reasoning_effort: Literal["minimal", "low", "medium", "high"] = Field(
"minimal", description="The reasoning effort to use when generating text reasoning models"
@@ -221,7 +209,7 @@ class OpenAIReasoning(BaseModel):
class OpenAIModelSettings(ModelSettings):
provider: Literal["openai"] = Field("openai", description="The provider of the model.")
- temperature: float = Field(0.7, ge=0.0, le=1.0, description="The temperature of the model.")
+ temperature: float = Field(0.7, description="The temperature of the model.")
reasoning: OpenAIReasoning = Field(OpenAIReasoning(reasoning_effort="high"), description="The reasoning configuration for the model.")
response_format: Optional[ResponseFormatUnion] = Field(None, description="The response format for the model.")
@@ -240,7 +228,6 @@ class OpenAIModelSettings(ModelSettings):
"max_tokens": self.max_output_tokens,
"reasoning_effort": self.reasoning.reasoning_effort,
"response_format": self.response_format,
- "parallel_tool_calls": self.parallel_tool_calls,
}
@@ -252,7 +239,7 @@ class OpenAIModelSettings(ModelSettings):
class AnthropicThinking(BaseModel):
type: Literal["enabled", "disabled"] = Field("enabled", description="The type of thinking to use.")
- budget_tokens: int = Field(1024, ge=0, le=1024, description="The maximum number of tokens the model can use for extended thinking.")
+ budget_tokens: int = Field(1024, description="The maximum number of tokens the model can use for extended thinking.")
class AnthropicModelSettings(ModelSettings):
@@ -279,18 +266,17 @@ class AnthropicModelSettings(ModelSettings):
"extended_thinking": self.thinking.type == "enabled",
"thinking_budget_tokens": self.thinking.budget_tokens,
"verbosity": self.verbosity,
- "parallel_tool_calls": self.parallel_tool_calls,
}
class GeminiThinkingConfig(BaseModel):
include_thoughts: bool = Field(True, description="Whether to include thoughts in the model's response.")
- thinking_budget: int = Field(1024, ge=0, le=1024, description="The thinking budget for the model.")
+ thinking_budget: int = Field(1024, description="The thinking budget for the model.")
class GoogleAIModelSettings(ModelSettings):
provider: Literal["google_ai"] = Field("google_ai", description="The provider of the model.")
- temperature: float = Field(0.7, ge=0.0, le=1.0, description="The temperature of the model.")
+ temperature: float = Field(0.7, description="The temperature of the model.")
thinking_config: GeminiThinkingConfig = Field(
GeminiThinkingConfig(include_thoughts=True, thinking_budget=1024), description="The thinking configuration for the model."
)
@@ -302,7 +288,6 @@ class GoogleAIModelSettings(ModelSettings):
"temperature": self.temperature,
"max_tokens": self.max_output_tokens,
"max_reasoning_tokens": self.thinking_config.thinking_budget if self.thinking_config.include_thoughts else 0,
- "parallel_tool_calls": self.parallel_tool_calls,
}
@@ -314,7 +299,7 @@ class AzureModelSettings(ModelSettings):
"""Azure OpenAI model configuration (OpenAI-compatible)."""
provider: Literal["azure"] = Field("azure", description="The provider of the model.")
- temperature: float = Field(0.7, ge=0.0, le=1.0, description="The temperature of the model.")
+ temperature: float = Field(0.7, description="The temperature of the model.")
response_format: Optional[ResponseFormatUnion] = Field(None, description="The response format for the model.")
def _to_legacy_config_params(self) -> dict:
@@ -322,7 +307,6 @@ class AzureModelSettings(ModelSettings):
"temperature": self.temperature,
"max_tokens": self.max_output_tokens,
"response_format": self.response_format,
- "parallel_tool_calls": self.parallel_tool_calls,
}
@@ -330,7 +314,7 @@ class XAIModelSettings(ModelSettings):
"""xAI model configuration (OpenAI-compatible)."""
provider: Literal["xai"] = Field("xai", description="The provider of the model.")
- temperature: float = Field(0.7, ge=0.0, le=1.0, description="The temperature of the model.")
+ temperature: float = Field(0.7, description="The temperature of the model.")
response_format: Optional[ResponseFormatUnion] = Field(None, description="The response format for the model.")
def _to_legacy_config_params(self) -> dict:
@@ -338,7 +322,6 @@ class XAIModelSettings(ModelSettings):
"temperature": self.temperature,
"max_tokens": self.max_output_tokens,
"response_format": self.response_format,
- "parallel_tool_calls": self.parallel_tool_calls,
}
@@ -346,7 +329,7 @@ class GroqModelSettings(ModelSettings):
"""Groq model configuration (OpenAI-compatible)."""
provider: Literal["groq"] = Field("groq", description="The provider of the model.")
- temperature: float = Field(0.7, ge=0.0, le=1.0, description="The temperature of the model.")
+ temperature: float = Field(0.7, description="The temperature of the model.")
response_format: Optional[ResponseFormatUnion] = Field(None, description="The response format for the model.")
def _to_legacy_config_params(self) -> dict:
@@ -361,7 +344,7 @@ class DeepseekModelSettings(ModelSettings):
"""Deepseek model configuration (OpenAI-compatible)."""
provider: Literal["deepseek"] = Field("deepseek", description="The provider of the model.")
- temperature: float = Field(0.7, ge=0.0, le=1.0, description="The temperature of the model.")
+ temperature: float = Field(0.7, description="The temperature of the model.")
response_format: Optional[ResponseFormatUnion] = Field(None, description="The response format for the model.")
def _to_legacy_config_params(self) -> dict:
@@ -369,7 +352,6 @@ class DeepseekModelSettings(ModelSettings):
"temperature": self.temperature,
"max_tokens": self.max_output_tokens,
"response_format": self.response_format,
- "parallel_tool_calls": self.parallel_tool_calls,
}
@@ -377,7 +359,7 @@ class TogetherModelSettings(ModelSettings):
"""Together AI model configuration (OpenAI-compatible)."""
provider: Literal["together"] = Field("together", description="The provider of the model.")
- temperature: float = Field(0.7, ge=0.0, le=1.0, description="The temperature of the model.")
+ temperature: float = Field(0.7, description="The temperature of the model.")
response_format: Optional[ResponseFormatUnion] = Field(None, description="The response format for the model.")
def _to_legacy_config_params(self) -> dict:
@@ -385,7 +367,6 @@ class TogetherModelSettings(ModelSettings):
"temperature": self.temperature,
"max_tokens": self.max_output_tokens,
"response_format": self.response_format,
- "parallel_tool_calls": self.parallel_tool_calls,
}
@@ -393,7 +374,7 @@ class BedrockModelSettings(ModelSettings):
"""AWS Bedrock model configuration."""
provider: Literal["bedrock"] = Field("bedrock", description="The provider of the model.")
- temperature: float = Field(0.7, ge=0.0, le=1.0, description="The temperature of the model.")
+ temperature: float = Field(0.7, description="The temperature of the model.")
response_format: Optional[ResponseFormatUnion] = Field(None, description="The response format for the model.")
def _to_legacy_config_params(self) -> dict:
@@ -401,7 +382,6 @@ class BedrockModelSettings(ModelSettings):
"temperature": self.temperature,
"max_tokens": self.max_output_tokens,
"response_format": self.response_format,
- "parallel_tool_calls": self.parallel_tool_calls,
}
diff --git a/letta/server/server.py b/letta/server/server.py
index be188dac7..f15e5c91e 100644
--- a/letta/server/server.py
+++ b/letta/server/server.py
@@ -436,8 +436,6 @@ class SyncServer(object):
handle = f"{request.model.provider}/{request.model.model}"
# TODO: figure out how to override various params
additional_config_params = request.model._to_legacy_config_params()
- additional_config_params["model"] = request.model.model
- additional_config_params["provider_name"] = request.model.provider
config_params = {
"handle": handle,
@@ -527,11 +525,6 @@ class SyncServer(object):
request.llm_config = await self.get_cached_llm_config_async(actor=actor, **config_params)
log_event(name="end get_cached_llm_config", attributes=config_params)
- # update with model_settings
- if request.model_settings is not None:
- update_llm_config_params = request.model_settings._to_legacy_config_params()
- request.llm_config.update(update_llm_config_params)
-
# Copy parallel_tool_calls from request to llm_config if provided
if request.parallel_tool_calls is not None:
if request.llm_config is None:
diff --git a/tests/configs/llm_model_configs/gemini-2.5-flash.json b/tests/configs/llm_model_configs/gemini-2.5-flash.json
index 51e58d2e6..387f1eb72 100644
--- a/tests/configs/llm_model_configs/gemini-2.5-flash.json
+++ b/tests/configs/llm_model_configs/gemini-2.5-flash.json
@@ -6,5 +6,5 @@
"model_wrapper": null,
"put_inner_thoughts_in_kwargs": true,
"enable_reasoner": true,
- "max_reasoning_tokens": 1000
+ "max_reasoning_tokens": 20000
}
diff --git a/tests/managers/test_agent_manager.py b/tests/managers/test_agent_manager.py
index 4c54a0cae..96056cc39 100644
--- a/tests/managers/test_agent_manager.py
+++ b/tests/managers/test_agent_manager.py
@@ -1250,7 +1250,7 @@ async def test_agent_state_schema_unchanged(server: SyncServer):
from letta.schemas.group import Group
from letta.schemas.llm_config import LLMConfig
from letta.schemas.memory import Memory
- from letta.schemas.model import ModelSettings
+ from letta.schemas.model import EmbeddingModelSettings, ModelSettings
from letta.schemas.response_format import ResponseFormatUnion
from letta.schemas.source import Source
from letta.schemas.tool import Tool
@@ -1271,10 +1271,9 @@ async def test_agent_state_schema_unchanged(server: SyncServer):
"agent_type": AgentType,
# LLM information
"llm_config": LLMConfig,
- "model": str,
- "embedding": str,
+ "model": ModelSettings,
+ "embedding": EmbeddingModelSettings,
"embedding_config": EmbeddingConfig,
- "model_settings": ModelSettings,
"response_format": (ResponseFormatUnion, type(None)),
# State fields
"description": (str, type(None)),
diff --git a/tests/sdk_v1/agents_test.py b/tests/sdk_v1/agents_test.py
index 2eee1c12e..6eb7c7dfc 100644
--- a/tests/sdk_v1/agents_test.py
+++ b/tests/sdk_v1/agents_test.py
@@ -5,9 +5,11 @@ AGENTS_CREATE_PARAMS = [
"caren_agent",
{"name": "caren", "model": "openai/gpt-4o-mini", "embedding": "openai/text-embedding-3-small"},
{
- # Verify model_settings is populated with config values
- # Note: The 'model' field itself is separate from model_settings
- "model_settings": {"max_output_tokens": 4096, "parallel_tool_calls": False}
+ # Verify model field contains the model name and settings
+ # Note: we override 'model' here since the input is a string but the output is a ModelSettings object
+ "model": {"model": "gpt-4o-mini", "max_output_tokens": 4096, "parallel_tool_calls": False},
+ # Note: we override 'embedding' here since it's currently not populated in AgentState (remains None)
+ "embedding": None,
},
None,
),
@@ -18,8 +20,8 @@ AGENTS_MODIFY_PARAMS = [
"caren_agent",
{"name": "caren_updated"},
{
- # After modifying just the name, model_settings should still be present
- "model_settings": {"max_output_tokens": 4096, "parallel_tool_calls": False}
+ # After modifying just the name, model field should still be present and unchanged
+ "model": {"model": "gpt-4o-mini", "max_output_tokens": 4096, "parallel_tool_calls": False}
},
None,
),