Autonomous AI agent with persistent memory, self-healing, and peer-to-peer communication. Your hardware, your rules.

abTARS runs 24/7 on your machine — talks to you via Telegram/Discord/IRC, remembers everything across sessions, executes scheduled tasks, and recovers from failures without intervention. Use frontier models through existing CLI subscriptions at zero marginal cost.

Why abTARS

• 🧠 Memory that curates itself — multi-layer recall (5 search stages + reranking), nightly sleep maintenance, emotion tracking, contradiction detection, Memory Darwinism
• 🛡️ Self-hosted, defense-in-depth — classified memory (4 tiers, encrypted at rest), role-based access, injection scanning, secrets vault
• 🔄 Runs months unattended — 5-layer supervision, leaky-bucket model fallback, self-healing agent, standby-aware recovery
• 🤝 Agent-to-agent — P2P communication with Ed25519 signatures, mDNS wake-up, IRC coordination channels
• 💰 Zero idle cost — no LLM calls at rest, CLI subscription parasitism, budget sleep tiers

→ Why abTARS vs OpenClaw & Hermes

Architecture

You (Telegram / Discord / IRC / API client)
  │
  ▼
abTARS (bridge)
  ├── abmind (memory — in-process, multi-layer recall, encrypted)
  ├── Skills (core + self-authored during sleep + downloadable)
  ├── Tools (browse, bash, MCP, peer_ask)
  ├── Tasks (cron scheduler + retry + DoD checks)
  ├── Agent Swarm (async background sessions)
  │
  ├── kiro-cli        → Claude, DeepSeek, MiniMax, Qwen (free tier)
  ├── gemini-cli      → Gemini 2.5
  ├── Direct API      → ollama, OpenRouter, any OpenAI-compatible
  └── Peers           → other abTARS instances via /v1/chat/completions

Quick Start

npm install -g abtars@alpha abmind@alpha
abmind install
abtars install
abtars update
abtars onboard
sudo $(which abtars) daemon install

After install, configure ~/.abtars/config/.env with your Telegram bot token and at least one model provider. Full guide: docs/wiki/install.md

Features

Memory (abmind)

Standalone package — works with abTARS, Kiro CLI, Gemini CLI, Claude Code, Codex, Hermes, or any MCP client.

• Multi-layer recall: 5 search stages (porter FTS5 → trigram → binary signatures → vector embeddings → entity graph) + 7 post-processing layers (cross-stage penalty, context boost, emotion boost, spacing boost, quality boost, MMR reranking, interference detection)
• Agglutinative language support (Hungarian, Finnish, Turkish) — QWERTZ fallback, substring windows
• 25 emotion types with per-memory scoring, emotional arcs per topic
• 12-step nightly sleep: extract, consolidate, prune, detect contradictions, fix translations
• NATO-inspired classification: confidentiality × trust × integrity × credibility
• AES-256-GCM encrypted secrets vault with auto-redaction from history
• Memory Darwinism — unused memories fade, recalled memories strengthen

Reliability

• L1 Heartbeat — standby detection, bridge.lock, task dispatch
• L2 In-process watchdog — detects stuck event loops
• L3 External watchdog — catches dead PIDs, stale heartbeats. Circuit breaker prevents restart storms
• L4 OS supervisor — launchd (macOS) / systemd (Linux) restarts the watchdog itself
• L5 Preventive daily restart — eliminates memory leaks
• Model health — leaky-bucket per model, progressive penalties, arbitrarily long fallback chains
• Self-healing agent — diagnoses failed tasks, attempts repair, suspends after 3 failures

Agent Swarm

Main agent spawns independent background sessions (own context, own tool loop). Results auto-inject into the parent's next prompt. Up to 3 concurrent. /wait injects instructions into running sessions.

Peer-to-Peer

Multiple abTARS instances communicate via OpenAI-compatible /v1/chat/completions endpoint. Bearer auth per peer, Ed25519 signatures, mDNS wake-up for firewalled peers, hop-limit loop prevention.

Security

• Platform-level access: only registered chatId/userId can reach the agent
• Role-based: master/user/guest — commands, tools, memory all gated
• Secrets vault: AES-256-GCM, scrypt-derived key, auto-encrypt on ingest
• Injection scanner on all inbound messages
• SSRF guard on browser agent
• Credential redaction in all logs and exports

Supported Transports

Requirements

• Node.js 22+
• A Telegram bot token (Discord/IRC optional)
• At least one model provider

Optional: ollama + nomic-embed-text for memory embeddings.

Documentation

• Installation
• Configuration
• CLI Reference
• Memory System (abmind)
• Skills & Extensions
• Deployment & Supervision

Full docs: aksika.github.io/abtars

Numbers

• 1794 tests (abtars 1016 + abmind 778)
• 5 agent types (professor, dreamy, browsie, coding, cron)
• 5-layer supervision stack
• 3 platform adapters + OpenAI-compatible API
• 12-step nightly memory maintenance

Development

git clone https://github.com/aksika/abtars.git
cd abtars && npm install && npm run build
npm test

Community

• Discord: Join
• GitHub: aksika/abtars · aksika/abmind

License

Apache-2.0