greatapi 2.0.0

The batteries-included FastAPI framework. Admin, auth, migrations, jobs and streaming out of the box — so you ship the backend instead of the plumbing.

FastAPI gives you a superb routing and validation layer and then stops. Every team builds the same next thing by hand: an admin, a login, Alembic wiring, a job queue, API keys. GreatAPI is that layer, and it is the same one whether you are serving a CRUD API or an LLM.

pip install greatapi
greatapi startproject myapp && cd myapp
cp .env.example .env
greatapi migrate && greatapi createsuperuser
greatapi runserver

You now have an API on http://127.0.0.1:8000, OpenAPI docs at /docs, and a working admin at /admin.

# main.py
from greatapi import GreatAPI

app = GreatAPI(title="myapp", installed_apps=["blog"])

# blog/admin.py
from greatapi import admin
from blog.models import Post

@admin.register(Post)
class PostAdmin(admin.ModelAdmin):
    list_display = ("id", "title", "published", "created_at")
    search_fields = ("title", "body")

That is the whole registration. greatapi startapp blog writes the model, schemas, repository, router and admin file, and adds the app to INSTALLED_APPS, so it serves requests and appears in the admin without another edit.

What you get

AI usage — tokens, cost and latency, recorded on every call	Jobs — a durable queue with retries, right in the admin
API keys — scopes, rate limits and monthly spend caps	List views — search, sort and paginate any model you register

The AI extra

pip install "greatapi[ai]"

from greatapi import ai

@app.post("/chat")
async def chat(request: Request, body: ChatIn):
    return ai.sse(ai.stream("anthropic:claude-sonnet-5", body.prompt), request=request)

@ai.tool
async def lookup_order(order_id: int) -> dict:
    """Fetch an order by its id."""
    ...

agent = ai.Agent(model="anthropic:claude-sonnet-5", tools=[lookup_order], max_steps=8)

@app.post("/agent")
async def run(request: Request, body: ChatIn):
    return ai.sse(agent.stream(body.prompt), request=request)

Tool schemas come from your type hints, so there is one source of truth. Every call and every agent step is recorded with its model, tokens, cost, latency and outcome, and shows up at /admin/usage — an agent run opens as a trace rather than a number in a log line.

Models are addressed as provider:model:

Try it without signing up for anything:

git clone https://github.com/sahajrajmalla/greatapi && cd greatapi
make install && make demo

Installing

pip install greatapi              # the framework. no LLM dependency at all
pip install "greatapi[ai]"        # + every provider and the agent tooling
pip install "greatapi[anthropic]" # + one provider
pip install "greatapi[local]"     # + self-hosted models, httpx only
pip install "greatapi[postgres]"  # + asyncpg

The AI code ships in the base wheel; only third-party SDKs live in the extras. import greatapi.ai always works, and reaching for a provider you have not installed tells you exactly what to install rather than raising an ImportError from four frames down.

Commands

greatapi startproject <name>          # a project that runs
greatapi startapp <name>              # an app, registered for you
greatapi runserver --port 8000        # the dev server
greatapi makemigrations -m "message"  # a migration from your model changes
greatapi migrate                      # apply them
greatapi createsuperuser              # an admin account
greatapi routes                       # every route in the app
greatapi generate-secret              # a signing key

Configuration

Everything is read from the environment or .env, prefixed GREATAPI_:

GREATAPI_SECRET_KEY=...                                  # required; greatapi generate-secret
GREATAPI_DATABASE_URL=postgresql+asyncpg://u:p@host/db    # any async SQLAlchemy URL
GREATAPI_DEBUG=false

SECRET_KEY has no default. With DEBUG off, a missing one stops the app starting and tells you how to make one.

Documentation

https://greatapi.readthedocs.io — quickstart, tutorial, admin, migrations, jobs, API keys, AI providers, agents, deployment, and the full settings reference.

Upgrading from 1.x? See MIGRATION.md.

Contributing

git clone https://github.com/sahajrajmalla/greatapi && cd greatapi
make install
make check       # ruff, mypy, and the test suite
make demo        # see it work

Details in CONTRIBUTING.md. Issues and pull requests are welcome; good first issues are labelled.

Security

Please report vulnerabilities privately — see SECURITY.md.

Credits

Built by Sahaj Raj Malla.

Standing on FastAPI, Starlette, Pydantic, SQLAlchemy and Alembic.

MIT licensed.