New · Pilot
Digital Serhii.
My engineering memory, inside your Claude.
Your AI agent forgets what the project decided the moment a session ends. Digital Serhii gives it a dedicated, isolated memory built on GBRAIN — with my way of working as its profile. Decisions persist across sessions, machines and teammates.
$ claude mcp add --transport http serhii brain.ovsiienko.ch/mcp
✓ serhii connected · 19 tools
› Where do we store the cart?
⟲ gbrain_recall · project acme-shop
#2 fresh · cart in Redis, 7-day TTL — anonymous carts must not pile up
In Redis with a 7-day TTL — that is what the project decided (#2). Orders stay in PostgreSQL.
Works with
- Claude Code
- Codex
- MCP clients over HTTP
Requirement: your client supports MCP over HTTP with a custom header.
01 — How it works
Connected in three steps
No new tool, no new interface. Your agent works as before — it just remembers.
- 01
Your own memory
You get a dedicated, isolated GBRAIN instance with my working profile and a personal access key.
- 02
One command
Your agent adds the memory as an MCP server. A short block in CLAUDE.md or AGENTS.md sets the project key.
claude mcp add --transport http serhii https://brain.ovsiienko.ch/mcp \ --header "X-API-Key: gbt_…" - 03
Work as usual
Before working, the agent asks the project memory; afterwards it saves verified decisions. The next session — even on another machine — finds them again.
02 — What your agent gets
Memory with principles
Project memory
Decisions, pitfalls and their reasons — retrieved by meaning, with freshness and a trust level. Not the chat history, but what has proven itself.
My working profile
Principles from my practice: replace valuable state only with a verified way back, keep effort proportionate, keep implementation, verification and delivery apart.
One project, one memory
The project key lives in the repository. Every machine and every teammate writes into the same project — not into twenty copies.
Honest sources
The agent separates what memory says from what it infers, and reports back which entries actually helped.
03 — Why GBRAIN
Not another notepad
GBRAIN is the memory layer I build for my own AI agents and use every day. Digital Serhii is a dedicated instance of it for your team.
Common today
The limit
GBRAIN
Built-in agent memory
The limit · Per vendor and per machine, as flat text
GBRAIN · One memory for every agent — Claude, Codex and others share the same facts
Memory libraries
The limit · For teams building their own agent; recall mostly by similarity
GBRAIN · For ready-made agents, connected with one command; ranked by outcome
Document search (RAG)
The limit · Finds text, not decisions; no feedback
GBRAIN · Keeps decisions, pitfalls and reasons — and learns what actually helped
Every hit carries a trust level (fresh, aging, verify, pitfall, contested). Wrong facts are corrected instead of silently overwritten — with history.
04 — Boundaries
What it is not
Better clear than surprising.
Not a chatbot
The thinking is done by your own agent, with your model and your account. No language model runs on my side.
Only what is saved
Memory holds what your agent explicitly saves — no recordings, no files from your repository.
Separate from everyone else
Every client has its own instance with its own keys. Nothing flows to other clients or into my own memory. Daily encrypted backups.
Not the human Serhii
The profile is a draft (v1) of my way of working. It makes no commitments on my behalf and its text grants no access.
05 — Questions
Frequently asked
What does the pilot cost?
The pilot is meant for a few selected projects. We agree the terms directly, depending on team size and scope.
Who can see our project data?
Only your team, through your keys, and me as the operator of the instance. Other clients have separate instances; my own memory is fully separate.
Can we get our data back or have it deleted?
Yes. On request you get an export of your instance, or it is deleted together with its backups.
Do we need your other products?
No. Digital Serhii is a standalone project memory. Further tools can be enabled selectively later.
A pilot with a few seats
Write briefly what your team is working on and which agent you use. I reply within one business day.
Request pilot access