# PromptMatrix

> Enterprise Prompt CI/CD & Runtime Behavioral Control Plane for AI Swarms (LangGraph, CrewAI, OpenClaw). Register, version, evaluate, and hot-patch agent personas without redeploying code. MIT Licensed.

PromptMatrix is the enterprise prompt management and CI/CD governance engine for AI engineering teams and autonomous multi-agent systems. Pull your prompts out of hardcoded application strings into a governed, auditable, sub-15ms cached registry.

## Quick Reference

- [Landing Page](https://promptmatrix.site/)
- [Developer Documentation](https://promptmatrix.site/docs/)
- [OpenAPI Spec](https://promptmatrix.site/docs/openapi.json)
- [Full LLM Knowledge](https://promptmatrix.site/llms-full.txt)
- [Python SDK (PyPI)](https://pypi.org/project/promptmatrix-sdk/)
- [CLI / Server Engine (PyPI)](https://pypi.org/project/promptmatrix-ai/)

## Core Endpoints

- `GET /pm/serve/{key}`: Sub-15ms cached hot-path prompt delivery with variable interpolation.
- `POST /pm/serve/{key}/feedback`: Telemetry feedback (latency, tokens used, outcome).
- `GET /api/v1/prompts`: List all prompts in an environment.
- `POST /api/v1/prompts/{key}/versions`: Create a new draft version.
- `POST /api/v1/approvals/{version_id}/approve`: Gate approval flow (admin/engineer only).
- `POST /api/v1/evals/run`: Multi-judge and rule-based prompt quality evaluation.

## Framework Quickstart

### Python SDK (`promptmatrix-sdk`)
```python
from promptmatrix import PromptMatrix
pm = PromptMatrix(api_key="pm_live_xxxx")
prompt = pm.serve("assistant.system", user="Alice")
print(prompt.content)
```

### OpenClaw Swarm Hot-Patching
```python
import openclaw
from promptmatrix import PromptMatrix
pm = PromptMatrix(api_key="pm_live_xxxx")
agent = openclaw.Agent(
    name="Researcher",
    instructions=pm.serve("openclaw.researcher.system").content
)
```

### LangGraph Node Integration
```python
from promptmatrix import PromptMatrix
pm = PromptMatrix(api_key="pm_live_xxxx")
def node_fn(state):
    sys_prompt = pm.serve("langgraph.node.reasoning")
    return {"out": model.invoke(sys_prompt.content + "\n" + state["in"])}
```

## Pricing Tiers

- **Community Edition**: $0 (Free, self-hosted, SQLite, MIT Licensed).
- **Founder Tier**: $499 one-time lifetime (100 org hardcap, unlimited prompts, 5 seats, 10k RPM).
- **Starter (Waitlist)**: $29/mo (100 prompts, 3 seats, 600 RPM).
- **Pro (Waitlist)**: $79/mo (400 prompts, 10 seats, 3k RPM, LLM evals).
- **Scale (Waitlist)**: $199/mo (1,000 prompts, 25 seats, 10k RPM).
- **Enterprise**: Custom ACV (SSO, dedicated infrastructure, 99.99% SLA).