STOP RENTING
YOUR INTELLIGENCE.
A local-first agent operating system. 19 LLMs. 18 channels. 145 tools. Zero telemetry.
Reddit Lead Hunter Pro
Find high-intent leads on Reddit before your competitors do. Fully local, zero scraping.
ONE CORE. MANY AGENTS.
Apache 2.0 core. Install the packs you need, when you need them. All run fully local on your machine. No subscription, no cloud.
Discord Lead Hunter
Monitor Discord channels with local LLM scoring. Free, bundled.
- Monitor Discord channels for messages matching your product. Scored by your local LLM. Requires COGNITHOR_DISCORD_TOKEN env var.
Hacker News Lead Hunter
Monitor HN top/new/best with local LLM scoring. Free, bundled.
- Monitor HN top, new, and best stories for posts matching your product. Scored by your local LLM.
Reddit Lead Hunter Pro
Find high-intent leads on Reddit before your competitors do. Fully local, zero scraping.
- Find high-intent leads on Reddit, score them with your local LLM, draft replies from 50 curated templates, and drop them into an encrypted CRM.
RSS Lead Hunter
Monitor RSS/Atom feeds with local LLM scoring. Free, bundled.
- Monitor RSS/Atom feeds for entries matching your product. Scored by your local LLM. No extra dependencies.
HOW IT WORKS.
Every request walks through three phases. No surprise tool calls, no silent data exfiltration, no LLM with unlimited keys.
Decomposes your intent into discrete tool calls using the local LLM you chose.
Verifies every tool call against GREEN / YELLOW / ORANGE / RED policies. Blocks risky stuff before it runs.
Runs the approved tools in a sandbox. Every action is audited to a SHA-256 hash chain.
WHAT'S INSIDE.
Everything a serious agent stack needs, preassembled. No glue code between model providers, memory backends and channel adapters.
Filesystem, shell, web, media, memory, vault, synthesis, code, skills, browser, arc, documents, vision, voice.
Local-first via Ollama, LM Studio, vLLM, llama.cpp. Cloud fallback for OpenAI, Anthropic, Google, Groq, Mistral.
Identity · Core · Knowledge · Episodic · Entities · Tactical.
Telegram · Discord · Slack · WhatsApp · iMessage · Matrix · CLI.
Screenshots, click, type, scroll, drag, vision element matching.
Autonomous learning, idle loop, meta-learner for self-improvement.
13/25 games solved, 4 solver families running in parallel.
Whisper + Piper. Push-to-talk or hands-free mode.
Memory + Vault + Web fusion with source confidence ratings.
GitHub-hosted, 5-check validation, publisher verified.
RECEIPTS, NOT VIBES.
BUILD YOUR OWN PACK.
KEEP 70%.
Coming Q4 2026. Creator marketplace with 70/30 revenue share. Publish once, earn forever. Your packs, your terms, your price.
Apache 2.0 core means anyone can write a pack. We handle payment, distribution, tax and license signing. You handle the ideas.
▶ Join waitlistLATEST WRITING.
All posts ▶Research updates, case studies, and opinion pieces from the Cognithor project.
Building a Reddit Lead Hunter Pack: 47 Sales in 30 Days
A case study of the first community-built Cognithor pack. How it was built in a weekend, what the first 30 days of sales looked like, and why this matters for the creator marketplace.
ARC-AGI-3: Teaching a Local Agent to Play Abstract Reasoning Games
We pointed a local Cognithor agent at ARC-AGI-3, the benchmark built to separate pattern-matchers from actual reasoners. We are two levels in. Here is what works, what does not, and what we learned.
Why Cognithor Is Local-First by Default
Every AI assistant you have ever used has sent your data somewhere you cannot see. The cloud-first default was chosen for the vendor, not for you. Here is what we built instead.
READY TO OWN
YOUR AI?
Cognithor is free. Install in three minutes. Runs fully offline with Ollama or LM Studio. No API keys, no subscription, no cloud round-trips.
Apache 2.0 · free for personal and commercial use · no telemetry