Promptsheon

Git-native version control for AI agents.

Promptsheon

Promptsheon is a self-hosted platform for shipping AI agents to production. It answers one question:

How do you author a multi-agent prompt, ship it safely, prove it doesn’t regress, and roll it back when it does?

You get a Fastify backend on :8080, a Next.js UI on :3000, a local SQLite file, and a content-addressed store. No cloud account, no signup, no telemetry. Run pnpm install && pnpm dev and you have a fully working prompt-management platform with multi-provider LLM, an audit chain, webhooks, eval scorers, and a live DAG editor.

Why Promptsheon?

Author prompts as DAGs. Drag-and-drop nodes for Planner, Agent, Tool, and Guardrail. Each node carries its own config and a live execution preview.
Ship with canary rollout. Versioned releases, per-environment activation, weighted traffic split, one-click rollback.
Make and check approvals. The release creator can’t approve their own release. Approvals persist with the reason and the voter.
Prove no regressions. Declarative eval suites, pluggable scorers (LLM-judge, regex, exact-match), parallel run results.
Self-evolve on regression. A scheduler watches live eval scores and re-plans the manifest when a release drifts.
Audit every action. Append-only, hash-linked audit log with a public verification endpoint.

Get started in 60 seconds

git clone https://github.com/sachncs/promptsheon.git
cd promptsheon
pnpm install
cp .env.example .env
pnpm dev

The backend listens on http://localhost:8080 and the UI on http://localhost:3000. No external services beyond the LLM provider you choose.

Continue with Getting started for a guided tour.

Architecture at a glance

┌────────────────────────────────────────────────────────────┐
│                        Browser                             │
│                  Next.js 16 + React 19                     │
└──────────────────┬─────────────────────────────────────────┘
                   │ /api/* (rewritten to :8080)
┌──────────────────▼─────────────────────────────────────────┐
│                       Fastify 5                            │
│ ┌────────────┬────────────┬────────────┬────────────────┐  │
│ │  Strands   │  Cedar     │   Audit    │    Repos /     │  │
│ │  Agents    │  Policy    │   Chain    │    CAS /       │  │
│ │  (Swarm,   │  Authorizer│  (hash-    │    Manifests   │  │
│ │   Graph)   │            │   linked)  │                │  │
│ └────────────┴────────────┴────────────┴────────────────┘  │
└──────────────────┬─────────────────────────────────────────┘
                   │
                   ▼
        ┌──────────────────────┐
        │  better-sqlite3 +    │
        │  content-addressed   │
        │  store on disk       │
        └──────────────────────┘

Every AI call goes through the Strands Agents SDK. Authorization is enforced by Cedar policies under packages/server/policies/. The audit chain is hash-linked and publicly verifiable at /api/audit/verify.

Where to go next

You want to…	Read this
Try it locally in 5 minutes	Getting started
Install on a server or container	Installation
Tune environment variables	Configuration
Understand the model	Core concepts
Hit the HTTP API	API reference
Read the architecture	Architecture
Walk through a guided build	Tutorials
Copy a working snippet	Recipes
Solve a specific problem	FAQ
Contribute code	Contributing
Report a security issue	Security

License

Promptsheon is licensed under the Apache License 2.0.