dbx-tools-graphiti
Run Graphiti REST, MCP, model routing, and PostgreSQL-backed graph memory from one Python runtime. The build synchronizes the upstream REST and MCP source plus the pinned PostGraph driver from Graphiti PR 1777 into the generated package tree, so the published wheel has no direct Git dependencies.
Both upstream surfaces come from the same Graphiti commit:
graph-servicefromserver/mcp-serverfrommcp_server/
Run
GRAPHITI_LISTEN=tcp://0.0.0.0:8000 \
uv run python -m dbx_tools.graphiti
From the repository root, the local launcher starts the unified runtime:
DATABRICKS_CONFIG_PROFILE=<profile> \
MODEL_NAME=databricks-gpt-5-nano \
bash scripts/run-graphiti-python.sh
The combined application exposes:
- FastAPI documentation at
/docs - OpenAPI at
/openapi.json - REST health and Graphiti routes, including
/healthcheck,/search, and ingestion endpoints - direct tool operations under
/tools/* - Streamable HTTP MCP at
/mcp/
Configure Graphiti through the shared environment contract used by the Node runtime and CLI.
The package also provides an environment-configured server entry point and a schema-only command:
python -m dbx_tools.graphiti
python -m dbx_tools.graphiti docs
docs writes the OpenAPI JSON document to stdout and exits without starting the
database runtime.
Databricks Model Resolution
The process creates one generated auth client and one generated model client at startup. It resolves the fuzzy chat model, ranks embedding endpoints by the configured name, embedding class, and dimensions, then resolves both endpoint routes. REST and MCP reuse the same Graphiti model clients, while the auth client injects refreshed Databricks headers into every model request.
Use From A Notebook
Use the same resolved option names without starting Uvicorn:
from dbx_tools.graphiti.runtime import GraphitiRuntime
async with GraphitiRuntime.from_options() as runtime:
graphiti = runtime.graphiti
In a Databricks notebook or Job, omitted auth options use the Python SDK's
default WorkspaceClient. Graphiti receives the runtime host and refreshed
authentication headers without copying credentials into environment variables.
Pass profile only when selecting an explicit profile instead.
The base package is suitable for Lakebase, external PostgreSQL, Spark notebooks, and serverless notebook Jobs:
uv add dbx-tools-graphiti
Graphiti depends on dbx-tools-node-runtime. A normal pip installation is
sufficient on Databricks serverless compute even when Node.js and npm are not
installed:
%pip install dbx-tools-graphiti
The shared runtime reuses system npm when available. Otherwise its first binding
load installs a compatible nodejs-wheel, creates direct build launchers, and
installs PythonMonkey into the active Python environment. Generated Graphiti
bindings import that runtime instead of carrying their own PythonMonkey loader
or Node shim copies.
For Python callers that omit databaseUrl, install the optional embedded
runtime:
uv add 'dbx-tools-graphiti[dev]'
Embedded mode starts bundled PostgreSQL with pgvector and persists it under
graphitiHome, or the platform data directory when no home is configured. The
owned PostgreSQL process stops with the runtime, while its data remains for the
next start. Node, CLI, and AppKit launchers select the extra automatically only
when no external database is configured.
Pass a regular PostgreSQL URL to use an existing database. A passwordless
non-local URL, Lakebase resource path, or Lakebase project name is resolved by
the generated Node Lakebase client. It injects a fresh short-lived credential
whenever the asyncpg pool opens a physical connection, without persisting that
credential in the URL. Lakebase mode provisions and uses the
databaseSchema schema, which defaults to dbx_tools_graphiti, because
application identities do not receive write access to public. Set
GRAPHITI_DATABASE_SCHEMA when multiple identities share one database. If the
database has no vector extension, Graphiti installs it in that schema. When
the database already has its single allowed vector installation, Graphiti
discovers that extension schema for type resolution without moving it.
Memory Writes
add_memory queues a write and returns immediately. Call
wait_for_memory_queue before stopping a short-lived runtime when the caller
needs queued work to be durable. Queue processing errors are re-raised by the
wait operation instead of being reported as an empty successful queue.
Use add_memory_sync when the request itself must wait for persistence. Both
forms remain available; queued writes are not forced to become synchronous.
get_queue_status reports the pending count and worker state.
When creating an episode, omit uuid and use the generated episode UUID from
subsequent retrieval. A supplied uuid selects an existing episode to update.
Composition Boundary
dbx_tools.graphiti.main adds FastAPI REST and MCP surfaces around the same
importable runtime. Shared environment options are parsed through generated
PythonMonkey bindings and mapped into both upstream settings objects.
1 maintainer
Metadata
1 maintainer
Release files for dbx-tools-graphiti 0.9.58
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| dbx_tools_graphiti-0.9.58.tar.gz | 540.1 kB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| dbx_tools_graphiti-0.9.58-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 1.1 MB
Release files / dbx_tools_graphiti-0.9.58.tar.gz
| Download URL | dbx_tools_graphiti-0.9.58.tar.gz |
|---|---|
| Size | 540.1 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
0d5c355385cb0973a7672b367cd7fdad6b805e929892f5afadb46b85865196d5
|
|
BLAKE2b-256 checksum How to use checksums |
af79ef7797cf05d4463b9c34e827da987e5257fb6084f68e4a00831de5adf008
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 7, 2026.
Transparency logRelease files / dbx_tools_graphiti-0.9.58-py3-none-any.whl
| Download URL | dbx_tools_graphiti-0.9.58-py3-none-any.whl |
|---|---|
| Size | 575.9 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
d0ce88af8ddc14172c48ae5b4711a57e4efe2417d07080af59d7f3ff3eac2ea0
|
|
BLAKE2b-256 checksum How to use checksums |
b87b5b6ea2ba56edea46d443280d04b57f3bf06c66701b9d89d5e2d899957f4b
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 7, 2026.
Transparency log
