One workspace your agents actually work in

Plan the day, run the meeting,write the doc, ship the change —and keep the thread.

BrainRouter is one workspace a whole team works in. Six things happen in it every week, and the useful part is not any one of them — it is that the sixth still knows what happened in the first.

A week, in one line

Work changes hands six times. It should not start over each time.

Each of these is a surface you can open on its own. What makes them a workspace is the line between them: the thing one hands to the next, still carrying where it came from.

  1. 01Plan

    The day, before anyone asks for a status update.

    Planner is the personal view of today across every project. Work you own on the team board arrives here as a mirrored item, so your day and the board never tell two different stories.

    PlannerTrack
    carries the shortlist you take into the room
  2. 02Meet

    A meeting that survives the meeting.

    Record and transcribe in the browser, then summarize through your organization's own model provider. Every meeting carries an explicit scope — private, team, organization, or a link that shares the redacted summary only.

    MeetingsTeams
    carries an action item, sent to Track in one click
  3. 03Write

    Where the decision gets written down.

    Notes holds pages, databases, and backlinks, with the same editor and the same keystrokes on the web and in the desktop app. One implementation, two places to reach it.

    NotesProjects
    carries a written decision the work can point back at
  4. 04Build

    The change gets made in the open.

    The agent workbench works inside a project's scope: the models you chose, the repositories you connected, the tools you permitted — and every step visible while it runs, not summarized afterwards.

    Agent workbenchRepositoriesAutomation
    carries a change, on its way to review
  5. 05Verify

    Nothing ships on confidence alone.

    Reviews read the diff and return findings that point back at code, tests, and named vulnerability intelligence. Assessments run only against targets someone authorized first.

    PR reviewsIssuesPentestsCVE intelligence
    carries a finding, with the evidence still attached
  6. 06Know

    And tomorrow, the thread is still there.

    The meeting summary becomes recallable with its transcript kept as the source underneath it. You can ask what the workspace knows, see why a result came back, and follow it to the material it came from.

    KnowledgeEvidenceRecall details

    and it carries back into the top of the next day

Keep the thread

Memory is not the product. It is why the other five survive each other.

A workspace that remembers everything and can explain nothing is a liability. So the record you get back can be taken apart: it names its source, its scope, its ranking, and the moment it changed.

  • Where did this come from?

    Records keep the files, commands, links, and transcripts they were built from.

    Evidence
  • Who is it for?

    Organization, team, project, and owner scope travel with the record instead of being applied at the end.

    Teams
  • Why did it come back?

    Ask a question and see which knowledge was chosen, how it ranked, and what put it there.

    Recall details
  • What if two things disagree?

    Conflicting records are surfaced as contradictions rather than quietly averaged.

    Contradictions
  • What changed, and when?

    Activity history follows when knowledge was saved, recalled, updated, or exported.

    Activity history
Four doors, one room

Change the window. Keep the workspace.

Models, connected accounts, project scope, permissions, knowledge, and review results do not turn into four separate products when you switch surface.

01The shared workspace

Dashboard

Planner, meetings, notes, the team board, agents, reviews, knowledge, and connections — the surface most of a team lives in.

02The deep workbench

Desktop

Chat, Code, and Track in one project shell, with files, terminal, tools, a browser, and reviews alongside the work.

03The same runtime, in a TTY

Terminal

The command-line workspace runs the same agent runtime, routing, policy, workflows, and knowledge as the apps.

04For the tools you already use

MCP + API

Governed tools and authenticated HTTP contracts, so editors, agents, and services can join the same workspace.

Boundaries

A shared workspace is only useful if sharing is a decision.

Meetings, notes, knowledge, and tools each carry a scope, and powerful actions stay behind an approval instead of being assumed.

Read the security policy
01

Actions stay explicit

Local execution, file changes, and sensitive tools stay behind the runtime's permission and approval policy.

02

Credentials stay behind the service

Connected accounts use server-sealed OAuth tokens. Provider secrets are write-only and are never returned to a client.

03

Sharing is a decision, not a default

Meetings, notes, and knowledge each carry a scope, and a public link shares a redacted summary rather than the recording.

04

Verification is evidence-led

Findings point back to code, diffs, tests, and attributable vulnerability intelligence rather than to model confidence.

Open and adaptable

Run the workspace on your terms.

The repository is MIT-licensed. Host it yourself, use the desktop app or the terminal locally, choose your model providers, and extend the runtime through MCP, hooks, connectors, skills, and agents.

Install the terminal workspace and the server$ npm install -g @kinqs/brainrouter-cli @kinqs/brainrouter-mcp-serverClone and inspect the system
Created and maintained byAnh DangBrainRouter creator and core maintainer