💎 Letta 1.0 Stable Release

[sarahwooders]

Letta (MemGPT) 1.0 Stable Release

Update 01/28/25: We expect only minor modifications to be made to our existing API, and will be announcing a 1.0 soon. Please see our updated Python and Typescript SDKs to use this the latest API.

We are working to release a stable version of MemGPT in a 1.0 release, which will be intended for production use. The release will include:

• Refactoring of DB code (e.g. the MetadataStore) to solve scaleability and connection issues (e.g. Add a consistent pool manager #1488)
• Consistent API interface for the REST API and the Python SDK, such as matching parameters and return types
• Introduce stable database schemas and alembic migrations for future releases
• Addition of production-grade testing, logging, and access control

To view progress on the release, you can check the integration branch.

General Cleanup

• Remove all nonessential code, dead or retired code, unofficially supported code from the codebase
• Isolate non-core elements into package add-ons
• Refactor MemGPT configuration to separate out server (e.g. DB) configuration and LLM/embedding configuration

Logging

• Singular parent logger throughout codebase and removal of all print and printd calls (except for in the CLI)
• Consistent use of py logging best practices for libraries and applications

Data Schemas

• Consistent data schemas (e.g. AgentState) across CLI, Python Client, and REST API represented with pydantic models
• Manage all persisted state (e.g. agent state) in an ORM
• Supported DB migrations for all future versions using alembic

Auth + Access

• Removal of authentication and access control (determined to be out-of-scope: developers should implement this within their own applications)

Testing

• Automated testing for all officially supported integrations (LLM/embedding provider, storage provider) - remove any untested integrations (suggest migrating to adapters or fork)
• SLA Support:
  • Atomic unit and integration tests (should not depend on a live OpenAI API key for testing)
  • Static code analysis threshold prospector or components of)
• Full CI run for each patch within all supported minors for Python client (versions 3.10, 3.11, 3.12)

Documentation

• Add auto-documentation feat: cleanup Python client and add auto-documentation  #1493

Timeline

Planned for end of January 2025:

• Database refactor and creation of new schema models
  • Support auto-migrations with Alembic
• refactor: move API to standardized pydantic schemas across CLI, Python client, REST server #1579
  • Server
  • Python Client
  • REST API
• REST API versioning feat: add V1 route refactor from integration branch into separate PR #1734
• Autodocs for Python SDK
• Refactor configuration
• Move integrations to external adapters / package add-ons
• Logging
• Release 1.0 API

[a67793581]

At present, I find that there is a problem with the generation of the open API. It cannot generate the correct parameters. Therefore, I can only manually modify the open API JSON file. The following is the /admin/users interface that I modified. It does not generate the body parameters for the GET request.

    "/admin/users": {
      "get": {
        "tags": [
          "admin"
        ],
        "summary": "Get All Users",
        "description": "Get a list of all users in the database",
        "operationId": "get_all_users_admin_users_get",
        "security": [
          {
            "HTTPBearer": []
          }
        ],
        "requestBody": {
          "description": "Parameters to filter users.",
          "required": true,
          "content": {
            "application/json": {
              "schema": {
                "$ref": "#/components/schemas/GetAllUsersRequest"
              }
            }
          }
        },
      "GetAllUsersRequest": {
        "type": "object",
        "properties": {
          "cursor": {
            "type": "string",
            "format": "uuid",
            "description": "Cursor to which to start the paginated request."
          },
          "limit": {
            "type": "integer",
            "description": "Maximum number of users to retrieve per page."
          }
        },
        "required": [],
        "example": {
          "cursor": "00000000-0000-0000-0000-000000000000",
          "limit": 50
        }
      },

[goetzrobin]

Outstanding API issues to get dev portal chat UI fully functional:

• feat: add endpoint to fetch agent by id #1639
• feat: implement detach data source from agent #1640
• feat: add available data sources endpoint #1641
• feat: add name of human and persona to memory API response #1642
• feat: add route that allows to attach multiple data sources to an agent #1643
• feat: add route that allows to detach multiple data sources to an agent #1644
• feat: add available tools endpoint #1645
• feat: add route that allows to add multiple tools at the same time #1646
• feat: add route to remove tool from agent #1647
• feat: augment /agents/{agentId}/archival/all route to support pagination/sort #1648
• feat: add endpoint to delete agent #1649
• feat: adjust agent response to include persona/human text and not just names #1650

[github-actions]

This issue is stale because it has been open for 30 days with no activity.

[Deng-Xian-Sheng]

Are there any plans to publish something like the OpenAI API so it can be easily integrated into any existing project?

[sarahwooders]

@Deng-Xian-Sheng yes we will be releasing documentation on our stable API release soon. You can see a preview here https://docs.letta.com/api-reference/agents/create