agentcookProduction-grade AI agent framework

Python core · Java business backend · TypeScript UI — everything you need to ship a real agent product.

🧠

9-dimension Harness

Memory · Tools · Context · Sandbox · Hooks · Routing · Telemetry · Compaction · Pruning — every dimension of a long-running agent, covered.

🌐

Bilingual Backend

Python runtime for agent core (FastAPI + asyncio) plus Java business layer (Spring Boot + DDD) — pick the right tool per concern.

🔌

Plugin Sandbox

Docker-based isolation for third-party tools. 5 attack-vector tests baked in. Capability-driven permission model.

📊

First-class Observability

OpenTelemetry traces, Prometheus metrics, Langfuse LLM telemetry, structured JSON logs — wired from day one.

🎨

Shared Design Tokens

Single source-of-truth for color, type, spacing — both admin (Vue 3) and chat app (React 19) consume the same tokens.

🚀

K8s + Helm Ready

Helm chart, blue/green workflow, multi-arch Docker images, Cloudflare Pages for the static surfaces.

What is agentcook?

agentcook is an opinionated agent harness that pulls together the moving parts every production agent needs — memory, tools, context management, sandbox isolation, hook pipelines, multi-model routing — behind a coherent API.

It is split across three runtimes by concern:

Python core (agentcook-core + agentcook FastAPI): the agent loop, memory layers, plugin runtime, model routing, observability.
Java business backend (agentcook-java): users, sessions, plugins, connectors, permissions — the operational data plane.
TypeScript front-ends (agentcook-admin Vue 3 + agentcook-app React 19 + Electron): admin dashboard and the end-user chat surface.

A Traefik gateway fans traffic between Python and Java, and a Helm chart deploys the lot to Kubernetes.

5-minute path

bash

git clone https://github.com/agentcook-cc/agentcook
cd agentcook-cc
make dev          # docker-compose: postgres, redis, jaeger, prometheus
uv sync           # python deps
uv run uvicorn agentcook_app.main:app --reload

Open http://localhost:8000/docs for the live OpenAPI explorer. Continue with the Quickstart to send your first chat request.