Replace hardcoded DEFAULT_MODEL and DEFAULT_SMALL_MODEL constants across all LLM clients with a centralized, configurable provider defaults system.
Key changes:
- Created provider_defaults.py with centralized configuration for all providers
- Added environment variable support for easy customization (e.g., GEMINI_DEFAULT_MODEL)
- Updated all LLM clients to use configurable defaults instead of hardcoded constants
- Made edge operations max_tokens configurable via EXTRACT_EDGES_MAX_TOKENS
- Updated cross-encoder reranker clients to use provider defaults
- Maintained full backward compatibility with existing configurations
This resolves the issue where Gemini's flash-lite model has location constraints in Vertex AI that differ from the regular flash model, and users couldn't easily override these without editing source code.
Environment variables now supported:
- {PROVIDER}_DEFAULT_MODEL
- {PROVIDER}_DEFAULT_SMALL_MODEL
- {PROVIDER}_DEFAULT_MAX_TOKENS
- {PROVIDER}_DEFAULT_TEMPERATURE
- {PROVIDER}_EXTRACT_EDGES_MAX_TOKENS
- EXTRACT_EDGES_MAX_TOKENS (global fallback)
Fixes#681
Co-authored-by: Daniel Chalef <danielchalef@users.noreply.github.com>
The cross_encoder for Gemini already supported passing in a custom client.
I replicated the same input pattern to embedder and llm_client.
The value is, you can support custom API endpoints and other options like below:
cross_encoder=GeminiRerankerClient(
client=genai.Client(
api_key=os.environ.get('GOOGLE_GENAI_API_KEY'),
http_options=types.HttpOptions(api_version='v1alpha')),
config=LLMConfig(
model="gemini-2.5-flash-lite-preview-06-17"
)
))
* add support for Gemini 2.5 model thinking budget
* allow adding thinking config to support current and future gemini models
* merge
* improve client; add reranker
* refactor: change type hint for gemini_messages to Any for flexibility
* refactor: update GeminiRerankerClient to use direct relevance scoring and improve ranking logic. Add tests
* fix fixtures
---------
Co-authored-by: realugbun <github.disorder751@passmail.net>
* migrate to pyright
* Refactor type checking to use Pyright, update dependencies, and clean up code.
- Replaced MyPy with Pyright in configuration files and CI workflows.
- Updated `pyproject.toml` and `uv.lock` to reflect new dependencies and versions.
- Adjusted type hints and fixed minor code issues across various modules for better compatibility with Pyright.
- Added new packages `backoff` and `posthog` to the project dependencies.
* Update CI workflows to install all extra dependencies for type checking and unit tests
* Update dependencies in uv.lock to replace MyPy with Pyright and add nodeenv package. Adjust type hinting in config.py for compatibility with Pyright.
* Implement telemetry feature for anonymous usage statistics collection in Graphiti; update Dockerfile CMD format for better signal handling; adjust Neo4j URI and healthcheck in docker-compose.yml; add new dependencies in pyproject.toml and poetry.lock.
* remove duplicated properties
* Update Dockerfile CMD to use JSON array format for improved signal handling
* remove tommlib dep only in 3.11
* Delete server/graph_service/logging_config.py
- Introduced `GroupIdValidationError` to handle invalid group ID formats.
- Added `validate_group_id` function to check that group IDs contain only alphanumeric characters, dashes, or underscores.
- Integrated `validate_group_id` checks in the `Graphiti` class to ensure group IDs are validated during processing.
- Introduced max_coroutines parameter in the Graphiti class to control the maximum number of concurrent operations.
- Updated the semaphore_gather function to accept max_coroutines as an optional argument, defaulting to SEMAPHORE_LIMIT if not provided.
- Adjusted multiple calls to semaphore_gather throughout the Graphiti class to utilize the new max_coroutines parameter for better concurrency management.
Fix trailing AND in edge_search_filter_query_constructor Cypher query
Corrected the edge_search_filter_query_constructor function to prevent
trailing AND operators in generated Cypher queries, which caused
Neo.ClientError.Statement.SyntaxError. Changed condition from
`j != len(and_filter_query) - 1` to `j != len(and_filters) - 1` for
valid_at, invalid_at, created_at, and expired_at filter blocks. Also
fixed outer loop condition to use `len(filters.<field>)` instead of
`len(or_list)`. Ensures valid Cypher syntax for single DateFilter cases.
Co-authored-by: TheEpTic <326774+TheEpTic@users.noreply.github.com>
* Refactor group_id handling and update dependencies
- Changed default behavior for `group_id` to 'default' instead of generating a UUID.
- Updated README to reflect the new default behavior for `--group-id`.
- Reformatted LLMConfig initialization for better readability.
- Bumped versions of several dependencies including `azure-core`, `azure-identity`, `certifi`, `charset-normalizer`, `sse-starlette`, and `typing-inspection`.
- Added `python-multipart` as a new dependency.
This update improves usability and ensures compatibility with the latest library versions.
* Update Graphiti MCP server instructions and refactor method names for clarity
- Revised the welcome message to enhance clarity about Graphiti's functionality.
- Renamed methods for better understanding: `add_episode` to `add_memory`, `search_nodes` to `search_memory_nodes`, `search_facts` to `search_memory_facts`, and updated related docstrings to reflect these changes.
- Updated references to "knowledge graph" to "graph memory" for consistency throughout the codebase.
* Update README for Graphiti MCP server configuration and integration with Claude Desktop
- Changed server name from "graphiti" to "graphiti-memory" in configuration examples for clarity.
- Added instructions for running the Graphiti MCP server using Docker.
- Included detailed steps for integrating Claude Desktop with the Graphiti MCP server, including optional installation of `mcp-remote`.
- Enhanced overall documentation to improve user experience and understanding of the setup process.
* Enhance error handling in GeminiEmbedder and GeminiClient
- Added checks to raise exceptions when no embeddings or response text are returned, improving robustness.
- Included type ignore comments for mypy compatibility in embed_content calls.
* Update graphiti_core/embedder/gemini.py
Co-authored-by: ellipsis-dev[bot] <65095814+ellipsis-dev[bot]@users.noreply.github.com>
* Update graphiti_core/llm_client/gemini_client.py
Co-authored-by: ellipsis-dev[bot] <65095814+ellipsis-dev[bot]@users.noreply.github.com>
---------
Co-authored-by: ellipsis-dev[bot] <65095814+ellipsis-dev[bot]@users.noreply.github.com>
