Episodic + semantic
Your agent stores moments and the meaning it consolidates from them — the way a brain does.
Memory that remembers.
A neurobiology-inspired memory layer for AI agents. Drop it in via MCP or API — your agent starts building real context in a single session.
Built on neurobiology
Not a database. A memory system.
Databases store. Brains remember — selectively, with context, associatively. BrainAI is the second one.
Drop-in via MCP
One JSON config and any MCP-compatible agent — Claude, Cursor, Zed — starts remembering.
Associative recall
Traces surface by context, not keyword. Ask sideways, find the right memory.
Decay & reinforcement
Low-signal memories fade on a power-law curve. High-signal ones consolidate into long-term.
Full inspection
A graph-native dashboard lets you audit, edit, and redact any memory — with lineage.
Yours, portable
Export the graph as JSONL or SQL. Self-host the MCP server. No lock-in.
How it works
Three steps to an agent that remembers.
01
Install the MCP server
One command — or self-host. Point your agent at it.
02
Your agent stores moments
Every exchange becomes an episodic trace, timestamped and tagged.
03
Recall happens associatively
Ask naturally. BrainAI walks the memory graph and returns what matters.
$ npx @brainai/mcp install $ export BRAIN_API_KEY=sk_brain_... # Agent ready.
Architecture
Six memory types, like a brain.
Working, episodic, semantic, procedural, personal, prospective. Same scaffolding the human brain uses — built into your agent.
WorkingWorking
What's in context right now.
EpisodicEpisodic
What happened — sessions, events, conversations.
SemanticSemantic
What you know — facts, entities, concepts.
ProceduralProcedural
What you can do — auto-skills from repeated patterns.
PersonalPersonal
Who you are — preferences, voice, identity.
ProspectiveProspective★
What you intend — future plans, time-tagged reminders.
* Theoretical foundation: Complementary Learning Systems (McClelland, 1995) and the CoALA cognitive architecture (Princeton, 2023). The first commercial implementation that ships a prospective layer.
For you
Your AI, on your terms.
01 · Switch mid-project
One conversation, every model.
Open Claude. Open ChatGPT. Open Cursor. The project is the same. Your context comes with you.
02 · Your defaults
Your assistant, not theirs.
It calls you by name. Knows you don't drink coffee after 3pm. Knows your music. Because you said so once.
03 · Yours to keep
Your memory, your file.
Export the graph as JSON, today. Self-host the server, today. No vendor owns your second brain.
Pricing
Pay for memories, not minutes.
Every plan includes the full memory model — episodic, semantic, procedural, working. Scale only what you use.
Hobby
Free
For solo builders and weekend agents.
- 10k recalls / month
- 1 agent
- 7-day memory horizon
- Community support
Most popular
Studio
$29/ mo
For devs shipping real agents.
- 500k recalls / month
- 10 agents
- Unlimited horizon
- MCP + REST + SDK
- Email support
Scale
Custom
For teams with agents in production.
- Unlimited recalls
- SSO + audit logs
- Self-hosted option
- SLA + Slack channel
- Dedicated engineer
Join the waitlist
Be early. Shape the product.
Closed alpha for 50 power-users in Q3 2026. Drop your email and tell us what you'd use it for.
Already 200+ engineers and writers from MIPT, Skolkovo and YC alumni.