Features
A memory that thinks along.
Hebby doesn't store loose notes. It understands connections, learns from every session and knows what your agent needs next.
Graph Memory
Thinks in connections. Not in notes.
At the heart of Hebby is a graph memory we built ourselves, tailored to AI agents. Projects, people, systems and decisions are linked together, just like knowledge in your head.
Classic AI memories find what sounds similar. Hebby finds what belongs together. Ask about a project and Hebby also knows the server behind it, the client involved and last week's decision.
Modeled on the brain
Four superpowers, borrowed from humans.
Built by psychologists who know how remembering works. And who turned it into technology.
Synaptic plasticity for your knowledge.
When neurons fire together, the brain strengthens the synaptic connection between them. Hebby applies that same core principle of neurobiology: it strengthens the links between memories that are needed together. A little sharper with every session.
Answers questions before they're asked.
Through associative networks, your brain activates knowledge before you consciously need it. Hebby does the same: it detects probabilistically where your work is heading and has the answer ready before your agent even has to ask.
Locks in knowledge like the brain during sleep.
At night, the brain moves experiences into long-term memory: it condenses, connects and sorts out. Hebby consolidates the same way. By morning, your memory is clearer than it was the evening before.
Stays current.
Outdated knowledge fades into the background, what matters stays present. Hebby relearns with you when projects, teams or plans change.
Complements your model
Your model knows the world. Hebby knows you.
Hebby doesn't duplicate what AI models can already do. It delivers exactly what no training dataset contains. That's why Hebby gets more valuable with every better model instead of becoming obsolete.
Future-proof
Models change at breathtaking speed. Hebby keeps taking on only what the model can't do itself, and follows every update automatically.
Scales effortlessly
Context files and built-in memory get more expensive with every line. Hebby stays lean, whether you have a hundred memories or a hundred thousand.
Token efficiency
Small investment. Many times back.
Measured in live operation across 144,000 API calls: depending on usage, Hebby adds 3 to 10 % in tokens, about 5 % on average. Save just two work steps per request and Hebby is already in the black.
Saves from the first prompt
Less research, fewer follow-up questions, fewer mistakes. Plus nearly empty context files like CLAUDE.md or AGENTS.md. It pays off even with small workloads.
Full cost control
Hebby measures itself. The built-in statistics show at any time what your memory costs and its share of your AI spend.
Gets cheaper and cheaper
Continuous consolidation condenses your memory. Less text, better hits, fewer tokens. Day after day.
For your agents
One second brain for all your agents.
Claude Code, Codex & MCP
Fully connected and part of every session automatically. Open to any MCP-capable agent.
Reranker on board. Rarely needed.
A lean reranker is built in. But Hebby learns what matters on its own. The longer it runs, the less it needs one. What is a reranker?
One memory, many agents
What one agent learns, all agents know. Across machines, tools and teams.
Permissions on your terms
Set who may know what, per agent and per person. Private stays private.
Portal with memory map
Inspect every memory and watch your memory grow, as a living map of your working world.
Made in Germany
Built by psychologists in Germany, with a high bar for data protection.
Early Access
Don’t worry, be Hebby.
An agent that remembers. Hebby is now rolling out to our first users. Tell us briefly what you work with.
