Build, deploy, and run autonomous AI agents 24/7.
India-first. Edge-native. Zero vendor lock-in.
GigaClaw is a self-hosted, autonomous AI agent platform. You deploy it to your own server or VPS, and it runs 24/7 — responding to messages, executing scheduled jobs, handling webhooks, writing code, managing files, and completing complex multi-step tasks.
It is built on a two-layer architecture:
- Event Handler — A Next.js server that handles real-time chat (web UI + Telegram), manages your agent's configuration, and creates jobs for the agent to execute.
- Agent Engine — A Docker container that runs your agent jobs using GitHub Actions or a local Docker daemon. The agent can write code, run shell commands, browse the web, and interact with GitHub.
GigaClaw is the only autonomous agent platform with native PragatiGPT support — India's indigenous Small Language Model for edge deployment, delivering 100% data privacy and zero foreign cloud dependency.
curl -fsSL https://raw.githubusercontent.com/gignaati/gigaclaw/main/install.sh | bashirm https://raw.githubusercontent.com/gignaati/gigaclaw/main/install.ps1 | iex# Create a new GigaClaw project
mkdir my-gigaclaw && cd my-gigaclaw
npx gigaclaw@latest init
# Then run the interactive setup wizard
npm run setupPrerequisites: Node.js 18+, Docker, Git
Step 1 — Create a new GitHub repository for your agent (e.g., my-gigaclaw).
Step 2 — Install GigaClaw into a local folder with the same name:
mkdir my-gigaclaw && cd my-gigaclaw
npx gigaclaw@latest init
npm installStep 3 — Run the setup wizard:
npm run setupThe wizard will ask for your setup mode:
- Hybrid (recommended) — Cloud + Local AI with smart routing
- Cloud — GitHub + ngrok + Telegram, full features
- Local — Ollama only, 100% offline
Step 4 — Start your agent:
docker compose up -dStep 5 — Chat with your agent at your APP_URL.
GigaClaw supports 6 LLM providers — more than any other self-hosted agent platform:
| Provider | Description | Data Privacy |
|---|---|---|
| PragatiGPT | Gignaati's India-first SLM — edge-native, on-premise | 100% — no foreign cloud |
| Ollama | Run any open-source model locally (Llama, Mistral, Qwen, Phi) | 100% — fully local |
| Claude (Anthropic) | claude-opus-4, claude-sonnet-4, claude-haiku-4 | Anthropic's servers |
| GPT (OpenAI) | gpt-5.2, gpt-4o, o4-mini | OpenAI's servers |
| Gemini (Google) | gemini-3.1-pro, gemini-2.5-flash | Google's servers |
| Custom API | Any OpenAI-compatible endpoint (vLLM, LM Studio, Together AI) | Depends on endpoint |
Set your provider in .env:
LLM_PROVIDER=pragatigpt # India-first, edge-native
LLM_PROVIDER=ollama # Fully local, zero cloud
LLM_PROVIDER=anthropic # Claude (default)
LLM_PROVIDER=openai # GPT
LLM_PROVIDER=google # Gemini
LLM_PROVIDER=custom # Any OpenAI-compatible API- Web Chat — Chat with your agent at your APP_URL
- Telegram — Connect a Telegram bot with
npm run setup-telegram - Scheduled Jobs — Cron-based recurring tasks via
config/CRONS.json - Webhook Triggers — POST to
/api/create-jobto trigger jobs programmatically - Code Workspace — Full terminal and code editor in the browser
- File Uploads — Upload images, PDFs, and text files to the chat
- Code execution — Write and run code in any language
- Shell commands — Execute terminal commands
- Web search — Search the internet for up-to-date information
- GitHub integration — Create PRs, manage issues, push commits
- File system — Read, write, and manage files in the repository
- Docker Compose — One-command deployment with Traefik reverse proxy
- Auto SSL — Let's Encrypt certificates via Traefik
- GitHub Actions — Agent jobs run in isolated Docker containers
- Auto-merge — Agent can merge its own PRs after review
- Hot reload — Push to
maintriggers automatic rebuild and restart
- Hybrid Mode — Cloud + Local AI with smart per-task routing (v1.6.0)
- PragatiGPT — India's indigenous SLM for edge deployment
- Ollama — Run any open-source model with zero cloud dependency
- Multi-LLM routing — Different LLMs for chat vs. agent jobs
- Per-job LLM override — Specify
llm_providerandllm_modelper cron job
Run both cloud and local LLMs simultaneously. GigaClaw automatically routes each task to the best provider.
npm run setup # Choose "Hybrid Mode" (recommended)| Strategy | Best for |
|---|---|
| Auto | Smart routing — complex tasks go to cloud, simple ones stay local |
| Cost-Optimized | Minimize API costs — local by default, cloud only when needed |
| Quality-First | Best output quality — cloud by default, local for drafts |
| Privacy-First | Maximum data privacy — local by default, cloud only for complex tasks |
- Setup configures a cloud provider (Claude, GPT, Gemini, PragatiGPT) and a local provider (Ollama)
- Each message is scored for complexity and privacy sensitivity
- The task router picks the optimal provider based on your chosen strategy
- Ollama availability is auto-detected at runtime — no reconfiguration needed
# Example .env for hybrid mode
GIGACLAW_MODE=hybrid
LLM_PROVIDER=anthropic # Cloud (primary)
LLM_MODEL=claude-sonnet-4-6
LOCAL_LLM_PROVIDER=ollama # Local (secondary)
LOCAL_LLM_MODEL=llama3.2
HYBRID_ROUTING=auto # auto | cost-optimized | quality-first | privacy-firstnpx gigaclaw init # Scaffold or update project files
npx gigaclaw setup # Run interactive setup wizard
npx gigaclaw setup-telegram # Configure Telegram bot
npx gigaclaw upgrade [@beta|version] # Upgrade to latest version
npx gigaclaw reset-auth # Regenerate AUTH_SECRET
npx gigaclaw reset [file] # Restore a template file
npx gigaclaw diff [file] # Show differences vs. templates
npx gigaclaw set-agent-secret <KEY> [VALUE] # Set GitHub secret (AGENT_ prefix)
npx gigaclaw set-agent-llm-secret <KEY> [VALUE] # Set LLM secret (AGENT_LLM_ prefix)
npx gigaclaw set-var <KEY> [VALUE] # Set GitHub repository variableThese files in config/ define your agent's personality and behavior. They are yours to customize — GigaClaw will never overwrite them:
| File | Purpose |
|---|---|
SOUL.md |
Your agent's identity, personality, and values |
JOB_PLANNING.md |
How your agent plans and breaks down jobs |
JOB_AGENT.md |
Instructions for executing jobs |
CRONS.json |
Scheduled recurring jobs |
TRIGGERS.json |
Webhook trigger definitions |
HEARTBEAT.md |
Tasks for the periodic heartbeat cron |
npx gigaclaw upgrade # Latest stable
npx gigaclaw upgrade @beta # Latest beta
npx gigaclaw upgrade 1.2.72 # Specific versionGigaClaw runs on any Linux server with Docker. Recommended:
| Provider | Spec | Monthly Cost |
|---|---|---|
| Hetzner CX22 | 2 vCPU, 4 GB RAM | ~€4 |
| DigitalOcean Droplet | 2 vCPU, 4 GB RAM | ~$24 |
| AWS EC2 t3.small | 2 vCPU, 2 GB RAM | ~$15 |
| Your own hardware | Any Linux machine | ₹0 |
For local development, use ngrok to expose your machine:
ngrok http 80
# Then update APP_URL: npx gigaclaw set-var APP_URL https://your-url.ngrok.ioContributions are welcome! Please read CONTRIBUTING.md before submitting a pull request.
- GitHub Issues: github.com/gignaati/gigaclaw/issues
- Discussions: github.com/gignaati/gigaclaw/discussions
- Email: support@gignaati.com
- Website: www.gignaati.com
Built with care by Gignaati — India's Edge AI Ecosystem