Cognithor
Cognithor · Agent OS: Local-first autonomous agent operating system. 19 LLM providers, 18 channels, 145 MCP tools, 6-tier memory, Agent Packs marketplace, zero telemetry. Python 3.12+, Apache 2.0.
Last verified:
What is Cognithor?
Cognithor is a local-first, open-source agent operating system that runs AI agents entirely on your own machine. It connects 19 LLM providers to 145 MCP tools across 18 communication channels, with a 6-tier memory system and zero telemetry. Think of it as your private ChatGPT replacement that you fully own and control, licensed under Apache 2.0 for personal and commercial use [web:7][web:3].
The core architecture follows the PGE Trinity loop: a Planner decomposes your intent into discrete tool calls using your chosen local LLM, a Gatekeeper verifies every tool call against GREEN/YELLOW/ORANGE/RED risk policies before execution, and an Executor runs approved tools in a sandbox with every action audited to a SHA-256 hash chain [web:7][web:3]. It includes 145 MCP tools across 14 modules (filesystem, shell, web, media, memory, vault, code, vision, voice, browser, ARC, documents), 18 built-in channels (Telegram, Discord, Slack, WhatsApp, iMessage, Matrix, CLI), and computer use capabilities with screenshots, click, type, scroll, drag, and vision element matching [web:7][web:8].
Cognithor features a 6-tier memory stack distinguishing Identity, Core, Knowledge, Episodic, Entities, and Tactical storage, an Evolution Engine for autonomous learning and self-improvement during idle time, voice STT+TTS with Whisper and Piper, ARC-AGI-3 benchmark support with 13/25 games solved, and knowledge synthesis with memory + vault + web fusion [web:7][web:8]. It supports local LLMs via Ollama, LM Studio, vLLM, and llama.cpp, with cloud fallback options for OpenAI, Anthropic, Google, Groq, and Mistral [web:2][web:7].
The tool is ideal for developers, researchers, personal assistants, operations managers, and anyone who needs a private, self-improving AI agent to handle complex multi-step tasks without sending data to the cloud [web:2]. Optional paid Agent Packs extend capabilities with specialized skills like Deep Research Analyst (multi-hop web research with citations), Reddit Lead Hunter Pro, Discord Lead Hunter, and more, while the core remains free forever [web:7][web:11].
Cognithor pricing
Pricing model: Freemium
The Cognithor Core is free under Apache 2.0 license for personal and commercial use with no subscription, no API keys, and no telemetry [web:1][web:7]. Optional Agent Packs are one-time purchases: Deep Research Analyst costs €65 (discounted from €119), Reddit Lead Hunter Pro costs €75 (discounted from €129), Content Creator OS costs €79 (discounted from €139), and Personal CRM costs €89 (discounted from €149) [web:7]. Free bundled packs include Discord Lead Hunter, Hacker News Lead Hunter, and RSS Lead Hunter [web:7]. All packs include lifetime updates, 48h founder support, 14-day money back guarantee, and MIT/Apache licensed source [web:7]. A creator marketplace launching Q4 2026 will offer 70/30 revenue share for third-party pack creators [web:7][web:11].
Cognithor pros
- Free and open source under Apache 2.0 license
- Runs entirely locally with zero telemetry
- No API keys required for local operation
- No subscription fees ever for the core OS
- 145 MCP tools across 14 modules
- 18 built-in communication channels including Telegram, Discord, Slack
- 19 LLM providers supported (local and cloud)
- 6-tier memory system for sophisticated recall
- Full offline capability with local LLMs
- GDPR-compliant with encryption at rest
- PGE Trinity safety gate blocks risky tool calls
- Computer use with screenshots, click, type, scroll
- Evolution Engine for autonomous self-improvement
- 13,247+ tests with 89.4% code coverage
- Voice STT+TTS with Whisper and Piper
- ARC-AGI-3 benchmark with 13/25 games solved
- One-time pack purchases instead of subscriptions
- Windows, macOS, and Linux support
- Encrypted audit log of all agent actions
- Knowledge vault with RAG and source confidence ratings
Cognithor cons
- Requires 16 GB RAM minimum, 32 GB recommended
- GPU (12 GB+ NVIDIA or Apple Silicon) recommended for good speed
- 27B planner model requires ~18 GB download
- No official support contract from core team
- Independent project maintained alongside day job
- Community support only via Discord and GitHub
- Creator marketplace not available until Q4 2026
- Cloud models require opt-in configuration
- First-time setup takes ~15 minutes for model download
- Web search unavailable when offline
Frequently asked questions about Cognithor
What is Cognithor exactly?
Cognithor is a local-first agent operating system. You install it on your own machine, point it at a local LLM via Ollama, LM Studio, or llama.cpp, and it runs a full Planner → Gatekeeper → Executor loop against a library of 145 MCP tools. The closest analogy is Jarvis that runs on your own hardware with no cloud, no telemetry, and no subscription [web:11].
How much does it cost?
The core is free under Apache 2.0. Optional paid packs are one-time purchases (Reddit Lead Hunter Pro is $79). You never pay to run Cognithor itself and never pay a recurring fee [web:11].
What hardware do I actually need?
16 GB RAM minimum, 32 GB recommended. On 16 GB you use the smaller 7B planner model; on 32 GB you get the default 27B one with noticeably better output quality. An NVIDIA GPU (12 GB+) or Apple Silicon dramatically speeds things up but is not required—a modern Intel/AMD CPU runs the 7B model at usable speed [web:11].
Does it run on Windows, macOS, and Linux?
Yes—all three. Windows 11 is the primary development target, macOS 13+ and Ubuntu 22.04+ are tested on every release. The binary is platform-native, not a web-app-in-a-window [web:11].
Do you collect any data about me?
No. The software has zero telemetry—no analytics pings, no crash reports, no anonymized usage tracking. The marketing website has no cookies, no analytics script, and no third-party embeds [web:11].
Will my conversations be used to train a model?
Never. Your chat history, vault, and memory tiers never leave your machine. The project will never train models on your vault, chats, or memory—it is an architectural property of the system, not just a policy [web:11].
What happens if I turn off my internet connection?
Cognithor keeps working. The Planner, Gatekeeper, Executor, Memory, Tools, and Skills all run locally. You lose web search and any channel needing network (Telegram, Discord), but the assistant itself—voice, CLI, vision, vault, skills—is fully functional offline [web:11].
Which LLMs can I use?
Any model Ollama, LM Studio, or llama.cpp can serve. The default planner is qwen3:27b, the default coder is qwen2.5-coder, and the default vision model is llava. The Multi-LLM router assigns different models to different roles and lets you override any assignment [web:11].
What is a pack?
A pack is a bundle of skills, tools, and configs that makes Cognithor good at a specific job—hunt leads on Reddit, triage inbox, research a topic. Packs ship as signed bundles and install through a 5-check validation pipeline [web:11].
Can I use Cognithor in my company?
Yes. The core is Apache License 2.0—use it commercially, modify it, redistribute it, build products on top of it. There is no enterprise tier that unlocks features. The same binary runs the same way whether you are a solo developer or a 500-person company [web:11].