Conclave: Agent Society

Multi-agent AI system with 58 tools, 10 LLM providers, and a web + CLI interface.

---

Quick Start (2 minutes)

git clone https://github.com/truehannan/conclave-coagent.git
cd conclave-coagent
npm run setup
conclave

That's it. The dashboard launches, detects it's unconfigured, and walks you through setup.

Windows: If npm run setup fails at the npm link step, run it as Administrator. Or skip linking and run directly:

node cli/dist/index.js

To update later: conclave update (pulls latest), then npm run setup again.

---

What Is This?

Conclave is a self-hosted AI agent that runs on your machine or a $6 VPS. It has:

• Agent Society — Multiple AI agents (Orchestrator, Researcher, Executor, Reviewer, Observer) collaborate on complex tasks
• 58 Built-in Tools — Shell, files, HTTP, search, code execution, services, scheduling, APIs
• 10 LLM Providers — DigitalOcean, OpenAI, Anthropic, Google, NVIDIA, HuggingFace, OpenRouter, GitHub, Cloudflare, Qwen
• 3 Interfaces — CLI dashboard (TUI), Telegram bot, Web frontend
• Live Model Fetching — Models pulled from provider APIs in real-time
• Skills Marketplace — Install domain skills from ClawHub (@user/skill)
• Dangerous Op Approval — Agent pauses for /approve before rm -rf, deploys, etc.

---

Architecture

User → [CLI / Telegram / Web] → Agent Core → Agent Society
                                     │
                   ┌─────────────────┼─────────────────┐
                   │                 │                 │
              58 Tools          Memory (D1)       10 Providers
              (shell,files,     (encrypted,       (live model
               web,code)        cloud sync)        fetching)

Agent Society Flow:

1. User sends complex request
2. Orchestrator decomposes into plan
3. Researcher gathers info → Executor runs commands → Reviewer validates
4. Observer monitors for failures
5. Compiled result returned

Simple requests are handled directly (no delegation overhead).

---

Features

Feature	Description
Agent Society	Multi-agent orchestration with 5 specialized roles
58 Tools	Shell, files, HTTP, search, code exec, services, scheduling, APIs, Composio
10 Providers	Live model fetching, switch with one command
Web Frontend	React dashboard with streaming chat, provider management
CLI TUI	Nano-like terminal UI with / command menu
Telegram Bot	Full agent access from your phone
D1 Cloud Sync	Credentials, memory, sessions sync to Cloudflare D1
Skills	Install domain expertise from ClawHub
Approval Flow	Agent pauses before dangerous operations
Observer Agent	Validates all execution before responding

---

CLI Commands

conclave              # Launch TUI dashboard
conclave deploy       # Deploy (localhost / remote IP / domain / path)
conclave start        # Start agent service
conclave stop         # Stop agent
conclave status       # Check health
conclave logs         # View logs
conclave model <name> # Switch model
conclave models       # List available models
conclave update       # Pull latest + rebuild

Inside the TUI: Type / to see all commands with arrow-key navigation.

---

Docker

docker pull ghcr.io/truehannan/conclave:latest

docker run -d \
  --name conclave \
  -p 8000:8000 \
  -v conclave-data:/opt/conclave \
  -e INTERFACE_MODE=cli \
  ghcr.io/truehannan/conclave:latest

Or build from source:

git clone https://github.com/truehannan/conclave.git
cd conclave
docker build -t conclave .
docker run -d --name conclave -p 8000:8000 -v conclave-data:/opt/conclave conclave

Open http://localhost:8000 to access the web UI. The setup wizard guides you through provider configuration.

Environment variables:

Variable	Default	Description
INTERFACE_MODE	cli	cli, telegram, whatsapp, all
CONCLAVE_API_PORT	8000	API server port
OPENAI_API_KEY	—	DigitalOcean/OpenAI key
COMPOSIO_API_KEY	—	Composio key (1000+ integrations)
TELEGRAM_TOKEN	—	Telegram bot token

---

Deploy

conclave deploy

Options:

• Localhost — Installs deps, builds frontend, starts locally
• Remote IP — SCP + nginx + systemd on your VPS
• Domain — Auto-SSL with Let's Encrypt + nginx
• Path — Serve at http://ip/path

---

Providers

Provider	Auth	Models
DigitalOcean	API Key	Llama, Mistral, DeepSeek, Claude, GPT
OpenAI	API Key	GPT-4o, o1, o3
Anthropic	API Key	Claude Sonnet/Opus 4
Google	API Key	Gemini 2.5 Flash/Pro
NVIDIA	API Key	Llama 405B, Nemotron
HuggingFace	API Key	Llama, Qwen, DeepSeek
OpenRouter	API Key	100+ models
GitHub	API Key	GPT-4o, Llama
Cloudflare	API Key + Account ID	Workers AI models
Qwen	API Key	Qwen Max/Plus/Turbo

All models fetched live from provider APIs — not hardcoded.

---

Tech Stack

• Backend: Python 3.9+ (~3000 lines) — agent.py, agents.py, tools.py, memory.py, api_server.py
• CLI: Node.js 18+ — esbuild bundled, raw stdin TUI
• Frontend: React + Vite + Tailwind — dark mode, streaming chat
• API: FastAPI + SSE streaming
• Storage: SQLite (local) or Cloudflare D1 (cloud)
• Encryption: Fernet (credentials at rest)

---

License

Source Available — Non-Commercial. Free for personal use.

---

Made by @truehannan