AI Agent Workflow Platform for Teams Running Real Work
Mrrlin is an AI agent workflow platform for planning, assigning, reviewing, and verifying AI-assisted business and coding work.
Chat tools vs Mrrlin
The execution layer beats another blank chat window.
Chat tools make you re-send context and manage the work. Mrrlin keeps project memory and spends tokens deliberately.
Why it matters
Agents spread across teams faster than the rules for using them.
Mrrlin gives teams the workflow layer around agents: goals, specs, tasks, runs, inbox decisions, artifacts, and evidence.
Workflow
Follow one security page through every step of the workflow.
Lay out the workflow
Describe the job: “publish a security page that answers what prospects ask during procurement.” The Director drafts the plan: collect the recurring questions from sales notes, write the answers, add the page and its footer link in the site repository, run QA, and request the publish decision.
Grant only what each step needs
Collecting questions needs no outside access. The page change needs the repository, so the edit is made in a git worktree set apart from your checkout and ends as a pull request. Publishing needs the site, and that takes its own grant, which records whether the action can be reversed and logs each call made through it.
Claims face a review panel
Security answers are claims a buyer will rely on, so they go through consensus: reviewers on different models, none of them the author, must converge before the draft moves on. Weak answers are returned with notes, and the pull request carries its diff summary and test results.
Decide on the evidence
The publish decision reaches the inbox with the preview address, the QA results and every review note. Approving triggers the production deploy, which runs as a GitHub Actions workflow in your own repository; sending it back returns the work with your reasons attached, and the final answers are saved to the wiki.
Use cases
Where it fits.
Agent workflow design
An operations lead wants every new vendor contract summarized, compared with the company's standard terms and returned with open questions. Set up in Mrrlin, that becomes a plan with a spec, a routine model for the summary, consensus on the risk notes, and a person's approval before anything goes back to the vendor.
Cross-functional execution
Accepting payments in euros touches four teams: finance sets the price table, engineering changes checkout in code, support needs reply drafts for billing questions, and the billing help article must match. One plan holds all four streams, each routed to a suitable agent, with dependencies keeping the article in step with what shipped.
Review governance
Governance is configured, not remembered. The approval list decides which actions stop for a person: messages to real contacts, anything published or deployed, and claims a client will read. Positioning goes to consensus, chosen tasks get human review, and internal drafts and research carry on unprompted, so review effort lands where the risk is.
Evidence tracking
Each run can be audited after the fact. It notes its branch and base commit, which provider executed it, and what it left behind: command output, test results, its diff summary, its self-review, the URL it produced. When someone asks in June why the security page changed, the answer sits on the task.
Comparison
Agent sprawl on one side, a shared workflow layer on the other.
Keep reading
More on AI workflow automation.
Start from the AI workflow automation overview, or go deeper with the guides next to this one.
FAQ
Before you start.
What does an AI agent workflow platform add if we already use Claude Code or Codex?
Those agents do the execution. Mrrlin supplies the workflow around them: the goal, the spec, task state, review by a different model, the approval inbox and project memory. Install @mrrlin-dev/mcp from npm, and Claude Code or Codex can read and update Mrrlin tasks, plans, wiki pages, runs and inbox items directly from inside a session.
Where do the models run, and who holds the keys?
On your machine, and the keys stay with you. A local bridge runs the agents there, and the usage counts against your existing subscriptions, such as Claude Code or Codex CLI, instead of a metered API. Mrrlin does not hold any model credential and does not resell tokens. The workspace keeps the state that outlives each session; the model stays yours to swap.
How should a team roll out an AI agent workflow platform?
Start with one cross-functional workflow that already hurts, such as the security page or euro pricing, and run it end to end. Set the required approvals in Settings first, keep the deploy policy on preview, and review the run records together. Add the next team once people trust the evidence the first workflow produced.
Tell us the outcome you want AI to execute.
Share the workflow you want to automate. We’ll map the first Mrrlin run — plan, agent routing, review loops, and approval checkpoints.
No credit card · No migration · One goal