Skip to content
Install on GitHub

The shared memory between humans and AI agents

Trust your agents.

FairMind Brain links requirements, decisions, issues and documents to the code, with their history, and agents read them before they act.

What an agent writes stays a proposal until a person confirms it.

Install on GitHub

Works with Claude Code, through a plugin and an MCP server.

agentperson
DEC-0418 · decisionconfirmed
Server-side sessions with opaque tokens
why
Immediate revocation, after the March logout incident.
anchored to
services/auth/session.py lines 88-140 · exact anchor
applies to
Every service in the company.
supersedes
DEC-0212, JWT tokens (2024). Stays readable.
proposedby an agent, Claude Code session, 12 Sep.confirmedby a person on the team, 13 Sep.

The trust layer for AI agents in software development

The development cycle with people, agents and BrainSix phases in sequence, each with an agent at work: plan, design, build, test, deploy, maintain. In plan the agent reads a user story from the Brain and writes a requirement as a proposal. In design it reads the requirement and proposes a decision that supersedes the previous one. In build it reads its task, writes the code and updates the task as a proposal. Each proposal becomes confirmed when the person at the next step approves. Test and deploy work on the code without touching the Brain. In maintain the agent reads the decision behind the code to fix a bug and opens an issue, which stays a proposal until a person confirms it. What is learned goes back to plan.

Where the memory comes from

From work your team already does. What comes from a source enters confirmed; what an agent writes waits for a person.

  • GitHub pull requestsWith their reviews, imported on request.
  • Uploaded filesPDF, Word, PowerPoint, Markdown and text.
  • Issue tracker / task managerJira, Linear or Notion: requirements and tasks stay where the team already keeps them.
  • MonitoringIncidents from failed checks.
  • AgentsWhat they write while they work.

The code. Brain anchors to files and functions, and rechecks the anchor on every read. billing/vat.py::apply_vat

When you can trust an agent

When it works on information you can verify.

  • Doesn't break what it can't see

    Requirements and decisions point to the code they cover, and the agent reads them before changing it.

  • Doesn't revive rejected decisions

    A superseded decision stays visible, marked as replaced, beside its successor and the reason.

  • Doesn't change memory on its own

    Agents propose, a person on the team confirms.

  • Doesn't guess

    Every link to code has a confidence level, and Brain says when the code can no longer be found.

Where Brain is used

Developers use it from their coding agent, the rest of the team from Brain Studio and from the tools it already uses.

Developers

From Claude Code, with the open source fairmind-coding plugin and the MCP server. The agent asks you fewer questions: it finds the answers in Brain, in one fast, deterministic search across code, Jira and GitHub, with fewer tokens and better accuracy. What you do answer goes into Brain for the rest of the team.

The team

In Brain Studio they look after the memory: they confirm what agents propose and remove duplicates. From Jira, Linear or an assistant like Claude, ChatGPT or Copilot, they design requirements using what Brain remembers.

fairmind-codingFree Claude Code plugin, MIT-licensed.

FairMind services

See Brain at work

In 30 minutes, on a case from your team.

Install on GitHub