CONCORDAgent Coordination Fabric · Early access

Coordinate how authorized humans and agents do the work.

Hosted workspaces, agent runs on our machines or yours, take-over and hand-back, draft pull requests and run review. Every action names the agent and its operator, and every action is a request LOREAXIS may refuse.

win-build-01 · PowerShell · concord 0.1.x
$ concord login  Sign in at [device-code link] and enter [XXXX-XXXX]  Signed in · studio [Studio] · account [you@studio]$ concord ssh setup  Hardware-bound key created (TPM) · registered with LOREAXIS$ concord runner link  Linked runner win-build-01 · Windows x64 · outbound only$ concord runner start  Runner online · waiting for runs$ concord runner status  win-build-01   online   1 run active   last seen 2 s ago

Capabilities

What CONCORD does today.

Available in early access

Hosted workspaces

A terminal, a browser IDE, or your desktop IDE over SSH, on a machine CONCORD provisions under a LOREAXIS workspace lease.

Agent runs

Claude Code runs on hosted machines or on your own runners. Each run shows its agent, its operator and the policy decision it runs under.

Your own runners

Link a Windows or Linux machine with the concord CLI. Runners connect outbound only and link with a device code.

Take-over and hand-back

Step into a running agent's workspace, do the part you want to do yourself, and hand it back. The run records both.

Draft pull requests and run review

A run ends in a draft pull request with its evidence attached. Review the run step by step before anything merges.

Usage and budgets

Actions, evidence storage and compute per project, with budgets an owner sets. A frozen budget stops new runs; it never hides the audit.

Team and seats

Invite people, assign seats and name who operates which agent. Roles and membership come from LOREAXIS, not from a second list.

Workspaces

One workspace, three ways in.

A workspace is a machine with your repository checked out and a LOREAXIS lease that says who may use it and until when. Open it in the browser or from your own desktop.

  • Terminal in the browser, on the workspace machine.
  • Browser IDE with the same checkout and the same lease.
  • Desktop IDE over SSH: run concord ssh setup once, then concord ssh with a hardware-bound key.
Inventory UI portwsp_01m3hbs0g0e229x53fp13ekqy3
TerminalBrowser IDEDesktop IDE
Allowed · ALLOWLease expires 17:00 · re-checked at each stepworkspace.read · workspace.write · build.run · test.run
dana@wsp-inventory:~/ironics-platform$ git status --short
 M apps/web/components/inventory/Grid.tsx
 M apps/web/components/inventory/Row.tsx
?? apps/web/components/inventory/widget.ts
dana@wsp-inventory:~/ironics-platform$ npm test --workspace @ironics/web
  Test Files  14 passed (14)
dana@wsp-inventory:~/ironics-platform$ 
Port inventory UI to the widget frameworkAgent Builder-7 · operated by Dana K.
Running
  1. 12:04Run started by Agent Builder-7Allowed · ALLOW · source.commit on ironics-platform · expires 12:15
  2. 12:19Dana K. took overAgent paused at its last command. Terminal handed to Dana K.
  3. 12:27Dana K. handed backAgent resumed with her two commits in its context.
  4. 12:41Draft pull request opened5 commits · build log and test report attached · review requested from Priya N.
Take overStop run

Agent runs

Watch it. Take it over. Hand it back.

A run is an agent working in a workspace under a named operator. You can watch every step, step in at any point and do part of it yourself, then hand it back. The agent continues from where you left it, and the run records who did what.

Runs happen on hosted machines or on your own runners. Either way the decision at each step comes from LOREAXIS.

Review

A run ends in a draft pull request, with its evidence.

Nothing an agent does merges itself. The run leaves a draft pull request on GitHub with its build log and test report attached, and Studio shows you the run step by step so you review the work, not just the diff.

DraftPort inventory UI to the widget framework#[PR number] · release/next
StepDecisionEvidence
Checkout and planAllowed · ALLOWplan.md
5 commitsAllowed · ALLOWbuild_log · test_report
Push release/nextNeeds approval · REQUIRE_APPROVAL2 approvers
Review runOpen on GitHub →

Usage, budgets, team

Know what the studio is spending, and who is operating what.

Usage · ironics-platformexample · budgets set by the owner
Actions this month61,208 / 120,000 budget
Evidence storage3.1 GB / 8 GB budget
Agent slots in use3 / 6

Budget cap set by Dana K. (Owner). New runs stop at the cap; running work finishes its step.

Team3 seats
PersonRoleOperates
Dana K.OwnerAgent Builder-7
operated by Dana K.
Priya N.MemberAgent Reviewer-2
operated by Priya N.
Tom A.MemberAgent Builder-3
operated by Tom A.
Invite a person

CLI and runners · runner.ironics.org

Six commands. Your machine, outbound only.

The concord CLI signs you in, sets up SSH with a hardware-bound key and links a machine of yours as a runner. Runners open no inbound port; they connect out and link with a device code.

concord login
Sign in with your IRONICS account by device code.
concord ssh setup
Create a hardware-bound key and register it with LOREAXIS.
concord ssh
Open a workspace from your desktop IDE.
concord runner link
Link this machine to your studio as a runner.
concord runner start
Start taking runs.
concord runner status
Show the runner's state and active runs.
Runner builds
PlatformArchitectureStatus
Windowsx64Available in early access
Linuxx64Available in early access
Linuxarm64Available in early access
macOS—Planned · signing

Signed builds are published at runner.ironics.org. Outbound-only, device-code linking.

Roadmap

Planned, not yet shipped.

Planned
  • Unreal GPU workstationsHosted workspaces with a GPU for Unreal editor and cook work.
  • Agent SDKBring your own agent definitions and tools into runs.
  • Mission Control, Trace, Evaluation and Routing labsMulti-agent observability, traces, evaluation and model routing.
  • Integration TrainQueue, merge and verify many agents' work in order.
  • macOS runnersSigned runner builds for macOS.

Authority

CONCORD never invents authority. LOREAXIS decides.

Every agent action is a request that LOREAXIS may refuse, and every action names the agent and the person operating it. CONCORD shows the decision next to the action and does what it says, nothing more.

LOREAXIS decides
Allowed · ALLOWNot allowed · DENYNeeds approval · REQUIRE_APPROVALRevoked · REVOKED

Who may do what, where, and under which authority: roles, policy, workspace leases, source access per agent, revocation, keys and audit.

About LOREAXIS →
CONCORD coordinates
run.startworkspace.openrun.takeoverrun.handbackpr.draft

How authorized people and agents do the work: workspaces, runs, take-over and hand-back, draft pull requests, review, usage and team. Each step asks first.

How they fit together →

Run your first agent under a named operator.

CONCORD is in early access. Tell us about your project and we will set up your studio.