Local-first, autonomous agent operating system for personal AI assistance.
Cognition + Thor — Intelligence with Power
Everything an autonomous agent needs, running entirely on your machine.
Ollama, LM Studio, vLLM, llama-cpp-python (local), OpenAI, Anthropic, Gemini, Groq, DeepSeek, Mistral, Together, OpenRouter, xAI, Cerebras, GitHub Models, AWS Bedrock, Hugging Face, Moonshot. Auto-detects from API key.
CLI, Flutter Web UI, REST API, Telegram, Discord, Slack, WhatsApp, Signal, iMessage, Teams, Matrix, Google Chat, Mattermost, Feishu, IRC, Twitch, Voice (STT/TTS with wake word).
Core identity, episodic logs, semantic knowledge graph, procedural skills, working memory, tactical memory. 4-channel hybrid search: BM25 + vector + graph + hierarchical document reasoning.
Planner (LLM) → Gatekeeper (deterministic policy) → Executor (DAG-based parallel sandbox). No hallucinated tool calls — every action is policy-checked before execution.
Filesystem, shell, memory, web, browser, media, vault, synthesis, code, skills, social listening, kanban, documents, identity. A2A protocol for agent-to-agent delegation.
Deterministic Gatekeeper (4 risk levels), 4-level sandbox (bubblewrap/Docker/JobObject), SHA-256 audit chain, Fernet AES-256 encryption, API rate limiting, CORS hardening.
Sci-Fi UI with glassmorphism, neon glow effects, 5 section colors, morphing sidebar, hacker mode with Matrix rain, Robot Office animation with 8 robots, real-time dashboard.
Wake word ("Jarvis") with Levenshtein + phonetic matching, Piper TTS (emotional voices), Whisper STT, WebSocket streaming, markdown rendering, file upload.
Obsidian-compatible Markdown vault with YAML frontmatter, tags, backlinks. Knowledge synthesis with confidence ratings, contradiction detection, gap analysis, and lineage tracking.
Guided Evolution through Pattern Analysis. Collects execution traces, performs causal root-cause analysis (10 categories), generates targeted optimizations, auto-applies with rollback safety.
CognitioEngine with 7 Genesis Anchors, personality vector (6 dimensions), emotion shield, reality check, character crystallization, narrative self, existential layer, dream cycle.
Install, search, rate community skills from GitHub registry. Publisher verification with trust levels. agentskills.io SKILL.md format compatibility (Hermes interop).
Cognithor learns from every interaction. GEPA analyzes failures, finds root causes, and generates targeted improvements — automatically.
Record execution traces
Causal root-cause analysis
Targeted optimizations
Test & deploy or rollback
Watch 8 animated robots work in real-time. CPU, GPU, RAM, and system load visualized directly in the office scene.
CPU, GPU, RAM, LOAD bars on wall screen
Robots navigate around desks via corridors
Sun, moon, stars, ceiling lights adapt
Dog and cat with AI behavior
From clone to running agent in under 5 minutes.
# Clone & install git clone https://github.com/Alex8791-cyber/cognithor.git cd cognithor pip install -e ".[all,dev]" # Pull local models (Ollama) ollama pull qwen3:32b # Planner (20 GB VRAM) ollama pull qwen3:8b # Executor (6 GB VRAM) ollama pull qwen3-embedding:0.6b # Embeddings # Launch cognithor # Full mode cognithor --lite # 6 GB VRAM mode COGNITHOR · Agent OS v0.90.0 Backend + UI at http://localhost:8741