Features

Everything you need to run coding agents in production.

Specs, guardrails, approvals, automations and an audit trail, for any agent on any model. Every feature is in the free download.

Download free Try the live demo

01 · Plan

Specs are the source of truth.

Write specs in your repository, or keep the spec format you already use. Structrl can draft one from an issue. Agents plan from the acceptance criteria, and every run reports which criteria it covered.

  • Lives next to the code, versioned with it
  • Popular spec formats read as they are
  • Issue to spec in one click
See it in the live demo
CHK-12 · .structrl/specs/chk-12-apple-pay.mdApple Pay at checkout
In progress
  • AC-1 Offered only when the browser supports it and the domain is verified.
  • AC-2 Creates the same payment intent as a card payment.
  • AC-3 The shipping contact pre-fills the order address.
  • AC-4 A decline keeps the cart and shows the standard message.

02 · Govern

Harnesses: the rules every agent runs by.

A harness sets the container, the checks that must pass, the budget and time limits, and whether a human approves before a pull request opens. Define it once and every run follows it.

  • Required and optional checks
  • Cost, duration and iteration caps
  • Approval gates where you want them
See it in the live demo
Standard PRrev 4
$5.00max cost
45 minmax duration
6iterations
  • Unit testsRequired
  • LintRequired
  • TypecheckRequired
  • BuildOptional

Ask a human before opening the pull request

03 · Code

Any agent. Any model. Your keys.

Pick the coding agent and the model for each job: leading cloud models, or open models on your own GPUs. You bring the keys on every plan, so there's no markup on tokens, and you can switch whenever something better ships.

  • Popular coding agents, or your own
  • Any OpenAI-compatible model, cloud or local
  • No token markup, ever
See it in the live demo
  • Coding agentopencode · anthropic/claude-sonnet-5.5
  • Review agentcodex · anthropic/claude-opus-5.5
  • Test writeropencode · qwen3-coder:30b

04 · Test

Watch the work happen, step by step.

Each run plans, implements and verifies in a sandbox, and records every model call, tool call, code change and check as it goes. See what the agent is doing now, and why it did what it did.

  • Live status per stage
  • Check output, not just pass or fail
  • Retry or cancel at any point
See it in the live demo
Waiting for your approvalelapsed 6:42
  1. PlanningGenerated plan (4 tasks)
  2. Implementation3 files changed · +47 −1
  3. VerificationUnit tests passed · 142 tests
  4. ReviewWaiting for your approval

05 · Review

You approve. Then it ships.

When a run needs a decision it lands in your Inbox with the diff, the files changed and the check results. Approve and the pull request opens with all the evidence attached. Reject and the reason goes on the record.

  • Changes to review before the PR exists
  • Approve or reject with a comment
  • Email when a run needs your approval
See it in the live demo
CHK-12: Apple Pay at checkoutApproval
  • addedsrc/payments/apple-pay.ts+24 −0
  • modifiedsrc/checkout/payment-step.tsx+9 −1
  • addedsrc/payments/apple-pay.test.ts+14 −0

06 · Release

Automations that start the right work.

A labeled issue, a new pull request, a push or a failing build on main starts a workflow with the harness and agent you chose. Your backlog starts moving on its own.

  • Events from your code platform and issue tracker
  • Filters on labels, branches and paths
  • Turn rules on and off per project
See it in the live demo
  1. WhenGitHub issue is labeled agent:implement
  2. Runspec-to-pr with the Standard PR harness
  3. AsCoding agent on Claude via OpenRouter
  4. ThenAsk for approval, open the pull request

07 · Audit

Every agent, every change, on the record.

Every step is hash-chained evidence: change or delete one and every hash after it breaks. See what every agent did across a workspace, and export the trail for your auditors.

  • Tamper-evident hash chain per run
  • Workspace-wide activity log
  • JSON export anyone can check offline
See it in the live demo
  • #14Added Apple Pay to the payment stepCodeChange 9f2c41d0
  • #15Unit tests passedCheckCompleted 4be8a7c2
  • #16All acceptance criteria covered (5/5)Verification c0d93e15
  • #17Approved by Priya ShahApprovalDecided 71aa6f08

Tamper-evident: 17 steps, each sealed by the one before

08 · Improve

Know what your agents are worth.

Success rate, cost, tokens and duration per agent and per spec. Find the harness that works, the model that pays off, and the specs that keep failing.

  • Cost and tokens per run
  • Success rate by agent and spec
  • Daily trends
See it in the live demo
94%success rate
$0.62avg. cost per run
6:52avg. duration

Runs per day, last 14 days

Also included

Built for how teams actually work.

Self-hosted and air-gapped

The whole platform in Docker on your infrastructure. Free for one person; teams add a license key that's verified offline.

Runs on your laptop

The desktop app can run Structrl on your computer, with your local repositories and a local model. It works offline.

Integrations library

Code platforms, trackers, agents, models and MCP tools. Search what's supported in the integrations library.

Knowledge

ADRs, conventions and docs that agents draw context from on every run.

Email notifications

Hear about approvals, failed runs and new pull requests by email. Everyone picks what they get.

Command-line tool

Start and follow runs from CI or your terminal with the structrl CLI and an access token.

Your AI engineering team starts with one issue.

Download Structrl free, connect a repository and your own model, and run your first agent today.

Download free Try the live demo