Open Source

Your AI Agent.
Your Rules.

Multi-engine agentic framework. Run models locally with Ollama, hit any API, or delegate to CLI agents. Swap backends with a single environment variable. No lock-in. No telemetry.

Get Started Explore Engines ↓
pureclaw
Engines

Choose Your Backend

Seven engines across three tiers. Switch with one env var or a tap in Telegram. No vendor lock-in.

Tier 1 — Local-First (Free)
Tier 2 — API (Per-Token)
Tier 3 — CLI (Subscription)
ollama Local
Local inference
Any GGUF model — Qwen, LLaMA, Mistral, DeepSeek, Gemma, Phi. Your hardware, your data.
🔧 6 built-in tools · Free
anthropic_api API
Anthropic Messages API
Claude Sonnet, Opus, Haiku. Direct HTTP with streaming. No SDK bloat.
🔧 6 built-in tools · Per-token
gemini_api API
Google Gemini REST API
Gemini 2.5 Flash, Pro. Pure HTTP — stdlib urllib for sync, aiohttp for async.
🔧 6 built-in tools · Per-token
openai_compat API
OpenAI-Compatible
ChatGPT, DeepSeek, Kimi 2.5, GLM 5, Grok, Mistral, vLLM — any OpenAI-compatible endpoint.
🔧 6 built-in tools · Per-token
claude_code CLI
Claude Code CLI
Full agentic tool use. File editing, bash, search — the complete Claude Code experience.
⚡ Full agentic · Subscription
gemini_cli CLI
Gemini CLI
Google's CLI agent with full tool use. Agentic coding and system tasks.
⚡ Full agentic · Subscription
codex_cli CLI
Codex CLI
OpenAI's Codex CLI agent. Full agentic tool use for code and system ops.
⚡ Full agentic · Subscription
Features

More Than a Chatbot

Channels, autonomous observers, human-in-the-loop email, and instant data cards. One process, zero external cron.

💬

Telegram Channel

Full conversational interface with streaming responses, inline keyboards, voice input via Whisper, photo and document analysis, and live model switching.

/backend → tap to switch engines
📧

Email Channel

IMAP polling classifies incoming mail, generates draft replies, and sends them to Telegram for explicit approve/reject before sending.

Human-in-the-loop · 3 accounts
👁

Observers

Seven autonomous background tasks on cron: email digest, morning brief, node health, daily snippet, content review, follow-up nags, and git push hooks.

Internal scheduler · No crontab
⚡

Dispatcher

Instant structured data cards that bypass the engine for speed. Weather, crypto, gold, stocks, forex, train departures, and infrastructure status.

Pattern match → direct API → card
📋

Draft Queue

Outbound email lifecycle: classify → draft → Telegram approve/reject → send via Gmail API → auto-create follow-up tracker.

Never sends without your approval
⏰

Task Scheduler

User-defined scheduled tasks and reminders via /schedule and /remind commands. Stored in SQLite, survives restarts.

/remind 2h Check server logs
🔧

Built-in Tools

All Tier 1 and Tier 2 engines get 6 tools out of the box: bash, read, write, edit, glob, grep. No external dependencies.

Shared tool loop across all backends
🎯

Session Memory

Persistent memory injection across conversations. SQLite-backed sessions, drafts, follow-ups, tasks, and observer state.

Context that survives restarts
Security

Pure, Simple, and Secure

Your agent runs on your hardware. Your data stays on your machine. No cloud platform between you and your models.

✓ Single-User Auth

One authorized Telegram user ID. No shared access, no multi-tenant complexity, no attack surface.

✓ Zero Telemetry

Phones home to nobody. No analytics, no tracking, no usage reporting. Check the source — it's open.

✓ No Cloud Dependency

Ollama keeps everything local. API backends use direct HTTP. No managed platform between you and inference.

✓ Minimal Dependencies

Stdlib HTTP for API calls. No bloated SDK chains, no transitive dependency trees, no supply chain risk.

✓ Human-in-the-Loop

Email drafts require explicit Telegram approval before sending. The agent suggests — you decide.

✓ Your Hardware

Run on your own server, your own GPU, your own network. Not a managed platform. Not a SaaS product. Yours.

Get Started

Up and Running in 60 Seconds

Clone, configure, run. Or deploy as a systemd service for always-on operation.

bash
$ git clone https://github.com/puretensor/PureClaw.git $ cd PureClaw $ pip install -r requirements.txt $ cp .env.example .env # Edit .env: set TELEGRAM_BOT_TOKEN, AUTHORIZED_USER_ID # Choose your engine: ollama, anthropic_api, gemini_api, etc. $ python3 nexus.py