NeuroAgent AI — The FastAPI of AI Agents

pip install neuroagent-ai

from neuroagent import Agent

agent = Agent(provider="anthropic", model="claude-opus-4-8")
print(agent.run("Analyze our Q3 sales data and write a short summary."))

Why NeuroAgent AI

You get	Instead of
`Agent` class	Gluing 5+ vendor SDKs together
`provider="openai" \| "anthropic" \| "gemini" \| "groq"`	Vendor lock-in
`memory="sqlite"`, `rag=True`, `tools=[...]`	Custom infra per project
`Workflow`, `Team`, `DatabaseAgent`, `APIAgent`	Reinventing orchestration
`agent.stats()`, traces, `agent.serve()`	Bolt-on observability & servers

How it compares

Capability	NeuroAgent AI	Raw vendor SDKs	Typical agent framework
Provider-agnostic core	Built in	✗ one vendor	Partial, often leaky
Tool calling	Auto-schema from signature	Hand-written JSON schema	Usually auto-schema
Memory backends	Session/SQLite/Redis/Vector	DIY	Varies, often one backend
RAG	Built in (`rag=True`)	DIY (load/chunk/embed/store)	Usually a separate library
Database agents (NL→SQL)	Built in, read-only by default	DIY	Rare, or third-party
Multi-agent teams	Pipeline/router/debate/hierarchical	DIY	Common, but a parallel system
Workflow DAGs	Parallel steps, retries, resume	DIY	Sometimes
Observability & cost	Automatic spans + token→$ cost	DIY	Often bolt-on or missing
Enterprise security	RBAC, audit log, secrets, sandbox	DIY	Rare
HTTP deployment	`agent.serve()`, one line	DIY FastAPI app	Rare, or heavy
Base install footprint	No vendor SDKs, extras opt-in	One SDK	Often pulls in many deps

"Typical agent framework" generalizes across the landscape of multi-agent/orchestration libraries — capabilities and trade-offs vary by project; check current docs for any specific framework before deciding.

Agents & Tools

One typed, async-first Agent with auto-schema tool calling. Learn more →

Memory

Session, SQLite, Redis, and semantic vector recall. Learn more →

RAG

Load PDFs/DOCX/text and chat grounded in your documents. Learn more →

Database & API Agents

Natural language → SQL or any HTTP API, read-only by default. Learn more →

Workflows

DAG orchestration with parallelism, retries, and checkpoint resume. Learn more →

Multi-Agent Teams

Pipeline, router, debate, and hierarchical collaboration patterns. Learn more →

Observability & Cost

Every run traced and costed automatically — no setup. Learn more →

Enterprise Security

RBAC, audit logs, secret masking, and sandboxed tool execution. Learn more →

Deploy in One Line

agent.serve() turns any agent into an HTTP service. Learn more →

Quality & Benchmarks

Real coverage, lint, and type-check numbers — not marketing claims. Learn more →

FAQ

Does NeuroAgent AI lock me into one LLM vendor?

No. Agent talks to OpenAI, Anthropic, Gemini, and Groq through one LLMProvider contract — switch with a string (provider="anthropic") and no other code changes. You can also register a custom provider for any OpenAI-compatible endpoint.

Do I need an API key to try it?

No. The built-in echo provider runs fully offline (no network, no key) so you can explore the entire API — tools, memory, workflows — before spending a cent. See Getting Started.

How heavy is the base install?

Four dependencies: pydantic, pydantic-settings, httpx, structlog. No vendor LLM SDKs — OpenAI/Anthropic/Gemini/Groq are implemented directly over httpx. Everything else (RAG loaders, SQL, Redis, encryption, the HTTP server) is an opt-in extra. See Quality & Benchmarks.

Are database and API agents safe to point at production systems?

DatabaseAgent and APIAgent are read-only by default — writes require an explicit opt-in. Pair them with RBAC and AuditLog from the Enterprise Security primitives for production deployments.

Is there a Docker image?

Yes — ghcr.io/theamitchandra/neuroagent-ai is published on every release. See HTTP Server for the run command.