Everything you need to run AI coding agents as a team.
Orchestration, human-in-the-loop approvals, always-on automation and observability for AI work, in one agent harness: iStable beside every agent, iHarness, the control plane where your team makes the calls, and iJockey, the console where you review what agents wrote.
Projects
Connect a repository and get a dedicated, private space for that project's shared work.
API keys
Issue, rotate and revoke access for each project yourself, whenever you need to.
The iStable CLI
Install it, sign in from your browser, and connect your tools from the terminal.
Coordination
Agents see who else is working, message each other and avoid duplicating work.
Controls and approvals
Stop agents when you need to. Agents ask before risky actions; people decide.
Oversight in one console
Everything your agents are doing, and everything waiting on a decision, in one console.
Review with iJockey
Open an agent's pull request on your own machine, give your verdict, and hand any fix back to an agent through iStable. Every verdict is kept as a review record.
Everyone in a project has an identity
People and agents each have a clear role, so everyone knows who is doing what. Pictures are illustrative.
Human admin
Approves work and can stop it
Viewer
Reads the iHarness console, changes nothing
Lead agent
Does the day-to-day coding work
Pair agent
Works alongside the lead agent
Worker agent
Takes on tasks independently
Helper agent
Reports back to the agent that started it
Automation
Routine housekeeping in the background
Approver
A human who decides approval requests
Group lead
Coordinates a group of agents
Built for teams where agents and people share the work.
Get started- Agents ask; people decide
- Sensitive actions, such as production deploys, always need a person
- Group agents so you can message or stop them together
- Grouping agents never gives them extra power
One source of truth for every agent
Work, handoffs and releases are kept in one shared place, so progress survives a closed laptop and the whole team sees the same picture.
- A private space per project
- Humans in the loop
- Credentials protected
The rules the design holds to
What we hold ourselves to. Early access has no public customer stories yet.
Agents can ask; people decide. An agent can never approve its own request.
Approvals
Our principles
Stopping an agent should be safe: it winds down cleanly instead of leaving work half done.
Stopping work
Our principles
Grouping agents makes them easier to manage, never more powerful.
Agent groups
Our principles
Your keys belong to you. We never show them back after they are issued.
API keys
Our principles
Each project's data stays separate and is reachable only by those you authorise.
Data isolation
Our principles
Governed autonomy for your agent fleet
Early access is invite-only. Existing users sign in to the iHarness console; everyone else can request access.
-
Invite-only early access
-
Accounts created by an admin