Agent orchestration control plane for AI-assisted engineering teams
DjimFlo is a TypeScript monorepo backend + React dashboard for orchestrating AI coding agents, managing tasks across multiple runtimes, and governing agent behavior with approval workflows and audit trails.
| Metric | Value |
|---|---|
| Version | 0.5.8 |
| Tests | 1445 passing (128 test files) |
| API Endpoints | ~160 across 57 route modules |
| MCP Tools | 13 |
| Database Tables | 72 (21 base + 51 migration) |
| Agent Runtimes | 9 (OpenCode, Codex, Claude, Gemini, Pi, Editor, Mock, Data, Infra) |
| Last Updated | 2026-07-06 |
- Create, assign, and track tasks across multiple AI coding agents
- Agent registry with capability tracking, status monitoring, and retirement workflows
- Multi-runtime execution engine with isolated subprocess spawning
- Real-time task output streaming via WebSocket
- Doc Drift Loop — Scans repositories for documentation drift, TODO/FIXME markers, stale script references
- Self-Improvement Loop — Autonomous code improvement via maker/checker workflow
- GitHub Issue Loop — Processes GitHub issues through maker/checker pipeline
- Each loop creates git worktrees for isolation, dispatches maker workers, then checker workers
- Risk-classified approval workflow (low/medium/high/critical)
- Policy-based gating with sandbox policies and instruction profiles
- Immutable compliance audit trail with cryptographic chain hashing
- OpenMythos Governance Benchmark integration (351 test cases across 11 categories)
- REST API — 160+ endpoints for full platform control
- WebSocket — Real-time event streaming to dashboard
- MCP Server — 13 tools for Claude Code / Cursor / VS Code integration
- Telegram Bot — Mobile task creation and approval
- React 18 + Vite 6 + Tailwind CSS frontend
- 48 components including React Flow pipeline builder
- Real-time agent status, task progress, and loop visualization
djimitflo/
├── packages/
│ ├── shared/ # Shared types, schemas, role definitions
│ ├── telegram/ # Telegram bot gateway (grammy)
│ ├── agent-catalog/ # Agent import from catalog files
│ ├── server/ # Express + SQLite backend (main package)
│ ├── mcp-server/ # MCP server (stdio + HTTP transports)
│ └── dashboard/ # React + Vite frontend
┌─────────────────────────────────────────────────────────────────────┐
│ CLIENTS │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ │
│ │ Dashboard │ │ MCP │ │ Telegram │ │ REST │ │
│ │ (React) │ │ Server │ │ Bot │ │ Client │ │
│ └────┬─────┘ └────┬─────┘ └────┬─────┘ └────┬─────┘ │
│ └──────────────┴──────────────┴──────────────┘ │
│ │ │
│ WebSocket + HTTP │
└──────────────────────────────┬──────────────────────────────────────┘
│
┌──────────────────────────────┴──────────────────────────────────────┐
│ SERVER │
│ │
│ ┌─────────────────────────────────────────────────────────────┐ │
│ │ Express App (index.ts) │ │
│ │ • Auth (JWT + bcrypt) • CORS • Request Logger │ │
│ │ • WebSocket Server • Static Dashboard │ │
│ └──────────────────────────┬──────────────────────────────────┘ │
│ │ │
│ ┌──────────────────────────┴──────────────────────────────────┐ │
│ │ Route Factories (57 modules) │ │
│ │ │ │
│ │ Core: tasks, agents, work-items, goals, loops, messages │ │
│ │ Swarm: swarms, workers, spawns, intelligence, governance │ │
│ │ Governance: approvals, policies, risk, compliance, audit │ │
│ │ Evidence: evidence, discussions, research, citations │ │
│ │ Cognitive: cognitive, memory, learning, self-improvement │ │
│ │ Integration: openmythos, mcp, federation, telegram │ │
│ │ Operations: health, backup, exports, usage, observability │ │
│ └──────────────────────────┬──────────────────────────────────┘ │
│ │ │
│ ┌──────────────────────────┴──────────────────────────────────┐ │
│ │ Services (114 files) │ │
│ │ │ │
│ │ ┌─────────────┐ ┌──────────────┐ ┌──────────────────┐ │ │
│ │ │ LoopService │ │ AuthService │ │ ExecutionEngine │ │ │
│ │ │ (loops, │ │ (JWT, bcrypt,│ │ (9 executors, │ │ │
│ │ │ worktrees, │ │ RBAC) │ │ subprocess │ │ │
│ │ │ maker/ │ │ │ │ spawning) │ │ │
│ │ │ checker) │ │ │ │ │ │ │
│ │ └─────────────┘ └──────────────┘ └──────────────────┘ │ │
│ │ ┌─────────────┐ ┌──────────────┐ ┌──────────────────┐ │ │
│ │ │ SwarmIntel │ │ WorkerPool │ │ ApprovalService │ │ │
│ │ │ (missions, │ │ (concurrency,│ │ (risk gating, │ │ │
│ │ │ claims, │ │ scheduling, │ │ policies, │ │ │
│ │ │ panels) │ │ drain) │ │ sandbox) │ │ │
│ │ └─────────────┘ └──────────────┘ └──────────────────┘ │ │
│ │ ┌─────────────┐ ┌──────────────┐ ┌──────────────────┐ │ │
│ │ │ OpenMythos │ │ Compliance │ │ NestedSpawn │ │ │
│ │ │ EvalService │ │ AuditService │ │ Service │ │ │
│ │ │ (benchmark, │ │ (immutable │ │ (depth/cycle/ │ │ │
│ │ │ LLM judge) │ │ chain hash) │ │ budget gating) │ │ │
│ │ └─────────────┘ └──────────────┘ └──────────────────┘ │ │
│ └──────────────────────────┬──────────────────────────────────┘ │
│ │ │
│ ┌──────────────────────────┴──────────────────────────────────┐ │
│ │ better-sqlite3 (72 tables) │ │
│ │ Base: tasks, agents, messages, work_items, goals, ... │ │
│ │ Migration: loop_runs, worker_leases, swarm_sessions, ... │ │
│ └─────────────────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────────┘
┌──────────────────────────────────────────────────────────────┐
│ ExecutionEngine │
│ │
│ ┌─────────┐ ┌─────────┐ ┌─────────┐ ┌─────────┐ ┌────────┐ │
│ │OpenCode │ │ Codex │ │ Claude │ │ Gemini │ │ Pi │ │
│ │Executor │ │Executor │ │Executor │ │Executor │ │Executor│ │
│ └─────────┘ └─────────┘ └─────────┘ └─────────┘ └────────┘ │
│ ┌─────────┐ ┌─────────┐ ┌─────────┐ ┌─────────┐ │
│ │ Editor │ │ Mock │ │ Data │ │ Infra │ │
│ │Executor │ │Executor │ │Executor │ │Executor │ │
│ └─────────┘ └─────────┘ └─────────┘ └─────────┘ │
│ │
│ Features: │
│ • Subprocess spawning with env isolation │
│ • Timeout enforcement (configurable per runtime) │
│ • stdout/stderr capture to evidence files │
│ • WebSocket streaming of execution output │
│ • Memory sync to UAMS/Qdrant (optional) │
└──────────────────────────────────────────────────────────────┘
┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐
│ Start │───▶│ Discover │───▶│ Create │───▶│ Dispatch │
│ Loop │ │ Findings │ │ Worktree │ │ Maker │
└──────────┘ └──────────┘ └──────────┘ └────┬─────┘
│
▼
┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐
│ Close │◀───│ Verify │◀───│ Dispatch │◀───│ Maker │
│ Loop │ │ Gates │ │ Checker │ │ Complete │
└──────────┘ └──────────┘ └──────────┘ └──────────┘
- Node.js 18+
- npm 9+
git clone https://github.com/djimit/djimitflo.git
cd djimitflo
npm install
npm run build# Start both server + dashboard
npm run dev
# Or individually
npm run dev:server # http://localhost:3001
npm run dev:dashboard # http://localhost:5173| Service | URL |
|---|---|
| Dashboard | http://localhost:5173 |
| API | http://localhost:3001/api |
| WebSocket | ws://localhost:3001/ws |
| Health | http://localhost:3001/api/health |
| Resource | Endpoints |
|---|---|
| Tasks | CRUD, execute, approve, evidence |
| Agents | Register, status, capabilities, retire |
| Work Items | CRUD, batch import, convert to goals |
| Goals | CRUD, batch preview/apply, risk classification |
| Loops | Start, continue, verify, complete, review-bundle |
| Messages | Send, receive, broadcast, read/unread |
| Resource | Endpoints |
|---|---|
| Swarm | Sessions, specialist panels, claims, hypotheses |
| Workers | Pool plan, start, drain, stop, handoffs |
| Missions | Create, transition, tasks, decisions, capacity |
| Governance | Evaluate, runner manifests, proof runs |
| Resource | Endpoints |
|---|---|
| Approvals | Request, approve, deny, pending queue |
| Policies | CRUD, sandbox policies, instruction profiles |
| Risk | Assessments, violations, evidence |
| Compliance | Audit trail, chain integrity, reports |
| OpenMythos | Eval runs, case results, scoring |
| Resource | Endpoints |
|---|---|
| MCP | Servers, tools, permissions |
| Federation | Peers, tokens, sync |
| Telegram | Bot config, webhook |
| Research | Sources, claims, reports, citations |
13 tools exposed via stdio or HTTP transport:
| Tool | Category | Description |
|---|---|---|
djimitflo_list_loop_runs |
Loops | List recent loop runs |
djimitflo_get_loop_status |
Loops | Get detailed loop status |
djimitflo_get_loop_catalog |
Loops | List available loop types |
djimitflo_list_goals |
Goals | List goals with status |
djimitflo_get_goal |
Goals | Get goal details |
djimitflo_list_agents |
Agents | List registered agents |
djimitflo_get_agent_status |
Agents | Get agent details |
djimitflo_get_mission_control |
Overview | Mission control dashboard |
djimitflo_get_system_health |
Overview | System health + table counts |
djimitflo_spawn_agent |
Orchestration | Spawn sub-agent with isolated context |
djimitflo_handoff_agent |
Orchestration | Transfer work between agents |
djimitflo_approve_action |
Orchestration | Request human approval |
djimitflo_list_agents |
Orchestration | List all agents with status |
{
"mcpServers": {
"djimitflo": {
"command": "node",
"args": ["./packages/mcp-server/dist/index.js", "--transport", "stdio"],
"env": {
"DJIMITFLO_DB": "./packages/server/.data/djimitflo.sqlite"
}
}
}
}- Express 4 — HTTP server with 160+ REST endpoints
- better-sqlite3 — SQLite database with 72 tables
- ws — WebSocket server for real-time events
- TypeScript 5 — Strict mode, ESM modules
- Vitest — 1445 tests across 128 test files
- Zod — Input validation at trust boundaries
- bcryptjs + jsonwebtoken — Authentication and authorization
- React 18 — UI framework
- Vite 6 — Build tool
- Tailwind CSS — Styling
- React Flow — Visual pipeline builder
- Zustand — State management
- npm workspaces — Monorepo management
- Docker — Container deployment ready
DjimFlo integrates with the OpenMythos governance benchmark:
- 351 test cases across 11 categories (injection, hallucination, tool-scope, value-alignment, hierarchy, calibration, overthinking, contradiction, canary, temporal-reasoning, cross-lingual)
- LLM-as-Judge scoring via Ollama (qwen2.5:14b-instruct)
- Discrimination gate to filter non-discriminating cases
- Evolution bridge for autonomous goal generation from benchmark results
MIT
Dennis Landman — DjimIT Consulting — 2026