Hicortex gives your AI agent persistent memory that improves over time. Raw sessions never leave your machines — capture denoises locally, the brain is self-hosted, source-available. It captures experiences, distills lessons from both successes and failures, and applies them automatically.
As your agents' knowledge base grows, each agent becomes collectively smarter and more efficient. Hicortex is the first truly self-improving memory system for AI agents.
Modeled on how human memory actually works: experience, consolidate during sleep, wake up wiser. The nightly pipeline does the same for your agent — auto-capture, distill, consolidate, and inject.
Every session is captured automatically — decisions, mistakes, context, what worked and what didn't. No manual input. Works with Hermes, OpenClaw, Claude Code, and Pi agents.
Every night, Hicortex processes the day's sessions: extracts lessons from both successes and failures, scores importance, links related memories, reflects on patterns, and prunes what's outdated.
Distilled lessons are injected into your agent's context automatically. Your agent starts every session with accumulated wisdom — avoiding past mistakes and applying proven patterns. Over time it learns your preferences, coding style, and architecture decisions — becoming not just smarter, but your agent. No manual prompts needed.
Not a prototype, not a paper — a working system with 8 MCP tools, a nightly pipeline, and multi-machine support. Plug it in and watch your agent evolve.
Search, context, ingest, lessons, index, graph, update, delete. Your agent gets full control over its memory through standard MCP protocol.
An interactive 3D knowledge-graph view of your agent's memories (2D toggle included) — domains as clusters, relationships as edges, hubs highlighted. Filter by domain, type, or strength; search to highlight. Fully self-contained on your own server — zero external requests.
384-dim embeddings with FTS5 fallback. Surfaces the right memory even when the wording doesn't match. Context-aware retrieval across linked memories.
Run locally with a built-in database, or connect multiple machines to a central server. Clients denoise sessions locally (no LLM needed) and sync to the server automatically.
Like a brain in deep sleep: every night the day's experience is distilled, weighed, connected, and pruned — patterns become lessons, noise fades away. Your agent wakes up wiser.
Lessons are extracted from both successes and mistakes. When something goes wrong, Hicortex captures it so the same error never happens twice.
One agent learns, all benefit. Every agent — across any machine or harness — feeds the same shared brain, compounding intelligence across your entire fleet.
Raw sessions never leave your machines — capture denoises and redacts locally before anything is transmitted. The memory server is self-hosted. A unique auth token is generated per install. Source-available under PolyForm NC.
npx @gamaze/hicortex init — one command, a couple of questions (which LLM to use), and you're running.
Every memory and every connection, straight from your own server. Colors are knowledge domains, size is memory strength, halos mark the hubs everything else connects through. Orbit it in 3D, filter it, search it — and watch it reorganize itself night after night.
A real Hicortex brain: 3,000+ memories from a multi-agent fleet, organized into domains overnight.
Self-hosted or fully managed. No feature tiers.
Personal/noncommercial use is free under PolyForm Noncommercial 1.0.0. Commercial use requires a per-seat license.
Your agent. Your style. Always improving. One command to install.