Quality & Benchmarks

Real numbers from the test suite and tooling that gate every merge — not marketing claims. Reproduce any of these yourself: ruff check . && mypy src && pytest --cov=neuroagent.

93.9%test coverage

0mypy --strict errors

0ruff lint errors

4deps in base install

Quality gates (CI-enforced)

Gate	Tool	Threshold	Status
Formatting	`ruff format --check`	Zero diffs	Passing
Linting	`ruff check`	Zero violations	Passing
Types	`mypy --strict` on `src/`	Zero errors	Passing
Tests	`pytest` (offline, mocked providers via `respx`)	All green	Passing
Coverage	`pytest --cov`	≥ 85%	93.9%

Base install dependency footprint

The base install pulls in exactly 4 packages — no vendor LLM SDKs. OpenAI, Anthropic, Gemini, and Groq are all implemented directly over httpx (already a base dependency).

pydantic base

pydantic-settings base

httpx base

structlog base

Capability extras

Everything else is opt-in. Missing an extra raises MissingDependencyError naming the exact pip install fix — never a raw ImportError.

Extra	Adds	Unlocks
`[rag]`	`pypdf`, `python-docx`	PDF/DOCX document loading
`[sql]`	`sqlalchemy`	Postgres/MySQL/SQL Server/Oracle database agents
`[postgres]`	`[sql]` + `psycopg`, `pgvector`	Postgres-backed vector store
`[redis]`	`redis`	Distributed memory backend
`[security]`	`cryptography`	Encrypted secret storage
`[server]`	`fastapi`, `uvicorn`	`agent.serve()` HTTP deployment
`[all]`	Everything above	Every capability at once

Public API surface

Core (Agent, Settings, Message…) 5

Composition (Workflow, Team…) 6

Security (RBAC, Sandbox…) 4

Integrations 4

Exception types 11

All exported names are typed and mypy --strict clean; counts reflect neuroagent/__init__.py as of v0.1.2. See the full list on the API Reference page.

Previous← Examples