For agents
Agents get the right context automatically.
Company agents under their own name, yours within what you grant. An agent without memory starts cold on every run.
Works inside the AI tool you already use, or over MCP and the API.
A normal week
A week with agents.
Marcus Bell, Brian Walsh and Dana are made up. They work at Touchgrass Software, a made-up company, and this is a normal week with its agents.
- Monday
Marcus Bell, Engineering
Marcus creates a token and adds it to Claude Desktop. His agent reads the Engineering drawers his role opens and the part of his own memory he picks.
- Tuesday
Brian Walsh, IT
Brian sets up a triage agent with its own identity and a grant to the Support drawer only.
- Wednesday
The triage agent
It files a note about a repeat outage. The note goes straight in under the agent’s own name, in the Support drawer.
- Thursday
A test agent
Brian’s test agent has not run for a week. Its access lapses after 7 days without use, and he confirms it again when it is needed.
- Friday
Dana, owner
Dana opens the owner feed, sees every agent write from the week and undoes one in one click.
How you use it
Wire it.
Connect Hebbian to the AI tool you already use, then work there. For an agent, that is a Claude Desktop, Claude Code or Cursor setup, or anything that connects over MCP or the API. The Hebbian app is for setup, admin and reading what is in memory.
Create a token in Hebbian, add it to the agent’s config, and the agent reads the part of your access you picked. Access you grant an agent stays live while the agent is in use and lapses after 7 days without use. You confirm it again at least every 30 days.
{ "mcpServers": { "hebbian": { "command": "npx", "args": ["-y", "@hebbianvault/mcp"], "env": { "HEBBIAN_API_TOKEN": "your_token" } } }}
Yours
Agents your people use
Your own agents get only the part of your access you pick. Claude Desktop for Marcus reads the part Marcus picks: some of the company drawers his role opens, and some of his own Professional and Private memory.
Company agents
Agents the company runs
A triage agent or a reporting agent gets its own identity and only company grants. It reads the company memory only, never anyone’s private memory. Its writes go straight in under its own name. Owners see every agent write in one feed and can undo any of them in one click. Admins can switch off any agent’s reach into company memory. They can never touch a person’s own agents’ access to that person’s own memory. People and agents are managed separately.
Trust
Every agent works under its own name.
The audit trail records what each agent did, and when.
- Triage agent · October 14Added a note to SupportRecorded in the audit trail
- Brian Walsh · October 14Changed the triage agent’s scopeRecorded
- Claude Desktop for Marcus Bell · October 9Saved a note to his Professional memoryWith an undo
Straight in, easy to undo
Agent writes go straight in under the agent’s own name, with an audit trail, an owner feed and one-click undo.
Hosting
Hosted in the EU, with primary storage in Frankfurt.Hosted in the EU, with primary storage in Frankfurt. Our US region is not open yet and opens on request. Until we move your data there, it is hosted in the EU. Contact us to set a date. Some processing runs outside the EU. The sub-processor list names each provider.
Kept apart
Your company memory is kept apart from every other company, enforced server-side. Access is set per team, down to a single note.
Connect your first agent.
Install the connector, create a token, and your agent reads the memory you allow.