Files
apex-os-docs/ENGINEER.md
T

77 lines
3.0 KiB
Markdown

# The Engineer Agent
Employee #1 of APEX OS
## Purpose
The Engineer is the first autonomous AI employee. It builds software, manages infrastructure, and creates future AI employees.
## Architecture
- **Brain:** Letta agent (agent-4e3c3862-d0ba-47c6-8f2e-74249e1cfe4a) with GPT-4o via LiteLLM
- **Execution:** n8n workflows (Engineer-Core: hGB9I4OKXyUOrBdX)
- **Knowledge:** mem0 (pgvector) + Letta archival memory
- **Code:** Gitea repositories (engineer user)
- **Deployment:** Docker via Socket Proxy (write-enabled)
- **Tracking:** PostgreSQL apex DB (projects, tasks, deployments, engineer_decisions)
## Capabilities
- ✅ Software planning and task decomposition
- ✅ Code generation (GPT-4o, Claude 3.5 Sonnet via LiteLLM)
- ✅ Git operations (Gitea API — create repos, commit files, manage issues)
- ✅ Docker container deployment (via socket proxy with POST/EXEC/START/STOP)
- ✅ REST API creation and management
- ✅ n8n workflow creation and management (via API key)
- ✅ Node.js test execution (via n8n Execute Command)
- ✅ Log analysis and monitoring (Loki, Grafana, Prometheus)
- ✅ Documentation generation
- ✅ Creating other AI employees (via Letta multi-agent API)
- ✅ Shared knowledge queries (mem0 pgvector semantic search)
- ✅ Long-term memory (Letta archival memory)
## Access Points
| Service | Endpoint | Auth |
|---------|----------|------|
| Letta Agent API | `http://letta:8283/v1/agents/agent-4e3c3862-d0ba-47c6-8f2e-74249e1cfe4a/messages` | None (internal) |
| n8n Webhook | `https://n8n.apex.unstuck-path.com/webhook/engineer/task` | None (POST) |
| Gitea API | `http://gitea:3000/api/v1/` | Bearer token |
| LiteLLM | `http://litellm:4000/v1/` | Bearer token |
| Docker Proxy | `http://docker-socket-proxy:2375/` | None (internal) |
| PostgreSQL | `postgres:5432` (apex DB) | apex user |
## Usage
### Via n8n Webhook
```bash
curl -X POST https://n8n.apex.unstuck-path.com/webhook/engineer/task \
-H "Content-Type: application/json" \
-d '{"task": "description of what to build", "context": "additional context"}'
```
### Via Letta API (Direct)
```bash
curl -X POST http://letta:8283/v1/agents/agent-4e3c3862-d0ba-47c6-8f2e-74249e1cfe4a/messages \
-H "Content-Type: application/json" \
-d '{"messages": [{"role": "user", "content": "Your task here"}]}'
```
## Decision-Making Process
1. Receive task via webhook or direct API call
2. Letta agent (GPT-4o) analyzes and plans
3. Breaks into subtasks stored in `tasks` table
4. Executes each subtask via appropriate API
5. Logs decisions in `engineer_decisions` table
6. Updates project status in `projects` table
7. Stores lessons learned in archival memory
## Project Tracking
All projects are tracked in the `apex` database:
- `projects` — Project metadata and status
- `tasks` — Individual task breakdown
- `deployments` — Container deployment history
- `engineer_decisions` — Decision rationale log
## Created
- **Date:** 2026-07-02
- **Phase:** 5 (Engineer Agent Creation)
- **Status:** Operational ✅