Shared agent memory
Decisions, solutions and pitfalls persist across sessions and agents. Entries carry sources, history and trust signals; feedback and verified outcomes help useful experience surface again.
In active development · Pilot projects
GBRAIN
Agents change. Knowledge and work stay connected.
I am building GBRAIN as a shared foundation for AI agents: they retain project knowledge, find the right tools and carry out tasks in steps that can be checked. For software development — and for recurring work inside a business.
A project by Serhii Ovsiienko · Lucerne
A shared foundation
Your AI agents
- Claude Code
- Codex
- Gemini
GBRAIN
- 01Memory
- 02MCP tools
- 03Orchestration
Code · documents · business systems
Independent of any single agent session. Connections depend on the MCP client, available interfaces and configured permissions.
01 — The platform
Remember. Act. Work together.
A language model answers questions. GBRAIN connects its agents to the context and workflows of a project or business.
Tools through MCP
External MCP servers extend the available capabilities. GBRAIN can discover their tools, make them searchable and invoke them. A growing tool collection becomes available where needed, without wiring every connection into every agent separately.
Orchestration you can follow
Goals become tasks with owners, dependencies and checks. Agents can hand work over; progress, results and necessary revisions remain visible. People set the direction and review the outcomes.
02 — Use cases
For code and everyday business work
The foundation stays the same. Knowledge, tools and workflow are configured around the task.
Continue the project without explaining it all over again
The task
Develop a feature with decisions and mistakes from earlier sessions in view.
- 01
Retrieve context
The agent finds earlier architecture decisions, known problems and useful checks.
- 02
Coordinate the work
Implementation and review receive clear tasks. Isolated working copies keep changes separate.
- 03
Check the outcome
Tests and changes are reviewed. Confirmed findings inform the next task.
The result
Reviewable changes and project memory that other agents can build on.
I use and continue developing the engineering capabilities in my own projects.
03 — Open interfaces
One gateway to a growing world of tools
MCP is a shared protocol that lets agents use external capabilities. GBRAIN brings those connections together with memory and task coordination.
- Code & repositories
- Documents & knowledge
- Email & requests
- CRM & customers
- Calendars & tasks
- Industry software
Discover
Read a registered MCP server's tools, descriptions and input formats.
Select what fits
Find suitable capabilities for a task, even across a large registered tool collection.
Use them in the workflow
Connect calls and results to the work so later steps can build on them.
Examples of possible connections, not a catalogue of customer systems already integrated. Each MCP server's availability, permissions and suitability are checked before use; some systems may require a custom adapter.
04 — How it works
A goal and a workflow you can follow
- 01
Set the goal
A person describes the task and the expected outcome.
- 02
Gather context
Agents read relevant memory and approved sources.
- 03
Carry out the work
Suitable tools and agents handle clearly scoped steps.
- 04
Check the result
Checks, review and human decisions determine what gets accepted.
- 05
Keep the experience
Confirmed decisions and findings are available for the next task.
A specific profile
Where does Digital Serhii fit?
GBRAIN is the platform. Digital Serhii is one application built on it: my engineering profile and selected experience as a foundation for an agent. For your business, the focus is your own knowledge, rules and workflows.
Explore Digital Serhii05 — Working together
A specific workflow in your business
I am looking for suitable pilot projects: one recurring task, a small team and an outcome we can check. We start with the need and choose the technology from there.
01
Identify the workflow and sources
What takes time? Which rules and systems would the assistant need?
02
Set up a limited pilot
Agree on data access, approvals and retention. Start with approved or anonymised examples.
03
Measure the benefit
Compare handling time, revisions and errors. Scope and price are agreed before starting.
Discuss your use case
Business changes and external actions require suitable access rights and approvals. An AI review does not automatically replace a responsible person's decision.