01
// for your users
they stop repeating themselves.
Preferences, decisions, and context carry across sessions, repos, and time. The agent shows up already knowing them — the retention moment your team is chasing.
// memory infrastructure for agent products
give your agents the memory they're missing.
brain is a typed memory substrate that connects to your product through a single MCP surface. Skip the schema design, the retrieval tuning, the multi-tenant story — ship the memory layer your users feel.
// what you ship
// three things your product gets day one
memory your users feel, not memory your roadmap fights.
brain handles the substrate so your team focuses on what makes the agent worth using. The benefits land in three places: your users, your engineers, and your retention metric.
02
// for your engineers
ship memory in a week, not a quarter.
Drop in an MCP server. Skip the schema design, the retrieval tuning, the lifecycle math, the multi-tenant scaffolding. Get back to the parts only you can build.
03
// for your agent
context it didn't know to ask for.
Semantic retrieval over a thinking graph surfaces relevant prior thoughts unprompted. Less prompt engineering. More natural recall. Smarter answers without longer prompts.
// the substrate
// typed memory + a thinking graph
eight typed tables, connected by a graph, queried through one mcp surface.
The schema enforces structure (a feedback isn't a fact isn't a lesson). The graph enforces connection (related memories surface together). The retrieval layer handles freshness, decay, and contradiction so your agent doesn't have to think about any of it.
Your code calls capture, search, connect. brain handles the rest.
user
feedback
lesson
work-pattern
reference
attention
thinking-rec
project
// see it in action
// 3-minute demo
give an agent persistent memory in 90 seconds.
Watch a Claude Code session capture a preference, close, and recall it unprompted from a fresh session two weeks later. No rules file, no static config, no re-explaining.
// what you skip building
// the four walls every memory team hits
six engineer-months of infrastructure that isn't your differentiation.
Every team that builds memory in-house converges on the same four problems. brain solves them as a substrate so your team's time stays where it matters — on the agent, not the scaffolding.
01 / wall
schema
What is a memory? Preference vs. fact vs. lesson vs. work-pattern vs. reference. Bag-of-strings collapses; typed schema scales.
02 / wall
retrieval
Semantic search that doesn't hallucinate connections. Edge traversal across related thoughts. Ranking that knows what's stale.
03 / wall
lifecycle
Contradictions, decay, supersession, freshness. Yesterday's preference shouldn't outvote today's. Wrong memories shouldn't compound.
04 / wall
tenancy
Multi-tenant isolation, audit trails, regional residency, the privacy posture your enterprise customers will demand on day two.
// what's real today
// no smoke, no mirrors
what's shipping vs. what's roadmap.
Most pitches hide the gaps. We don't, because the design-partner conversation goes faster when both sides know what they're integrating against.
| capability | status | notes |
|---|---|---|
| typed schema (8 tables) | shipped | battle-tested in daily operator use |
| semantic search + thinking graph | shipped | vector + edge model, in production |
| MCP endpoint | shipped | brain.bfurg.com/mcp · two-tier auth |
| multi-tenant isolation | roadmap · q3 | design-partner pilots get dedicated tenants |
| contradiction detection | roadmap · q3 | schema supports it; logic in design |
| domain memory packs | roadmap · with partners | tuned to your vertical's vocabulary |
| SOC 2, audit log, region pinning | roadmap · enterprise tier | scoped after first paid partner |
// design partner program
// 2–3 partners · 8-week pilot · open now
shape the substrate the rest of the market gets locked into.
brain's schema is mature enough to integrate today, but malleable enough to bend to your domain. After commercial launch, the schema calcifies. Design partners ship first, pay nothing during the pilot, and influence what enterprise sees later.
// what you get
- free hosted access throughout the 8-week pilot
- dedicated tenant tuned to your domain
- direct integration support from the operator who built it
- lifetime preferential pricing on the commercial product
- optional co-marketing as a launch partner
- schema influence before the substrate calcifies
// what you give
- real integration into a product surface your users touch
- weekly 30-minute sync during the pilot
- honest feedback on what the schema gets wrong for your domain
- permission to anonymize learnings for the public product
- engineering time to wire the integration in
// process
- intro call20 minutes — confirm fit
- technical scope1 hour — map your surface to brain's schema
- integration sprint2 weeks — wire it in
- production pilot6 weeks — real users, weekly sync
- decision gateconvert, extend, or part as friends
your users shouldn't have to repeat themselves.
If the agent retention metric your team cares about is "did the user have to re-explain themselves" — brain is the layer you'd otherwise be building. We're picking 2–3 partners. Apply this month.
apply for design partner access →