ForgeLoopSign in with GitHub

Autonomous software delivery

Turn GitHub issues into
evidence-backed pull requests.

ForgeLoop is the control plane around coding agents: it delegates bounded work, runs it on your infrastructure, repairs failures, proves acceptance criteria, and delivers the reviewed commit.

The delivery loop

Agents write code. ForgeLoop establishes confidence.

Repository agnostic

Connect any GitHub repository through a least-privilege GitHub App installation.

Closed-loop delivery

Plan, implement, integrate, verify, repair, independently review, and deliver one exact commit.

Evidence, not vibes

Every gate produces checksummed evidence, criterion coverage, provider telemetry, and an audit trail.

Your runners

Source code, shell commands, containers, and browser tests stay on enrolled infrastructure you control.

Designed for real repositories

Explicit boundaries at every layer.

Signed GitHub webhooks, short-lived installation credentials, tenant-scoped membership, task leases, isolated worktrees, policy-selected verification, immutable evidence, bounded retries, and guarded expected-SHA delivery.

  • GitHub identity and persisted role membership
  • No repository source stored in the control plane
  • Runner-local secrets and MCP processes
  • Human approval and opt-in guarded auto-merge

From setup to deliberate execution

A clear boundary between connecting and running.

Authorize your repositories

GitHub sign-in identifies you. GitHub App installation grants repository access. Intake policy decides which issues become eligible for work.

Follow the workflow →

Keep execution on your runner

Save provider keys locally, verify prerequisites, and explicitly start processing. Setup checks do not call a model; work execution can incur charges.

Explore runner setup →

Review the result, not just the summary

Inspect the diff, acceptance coverage, gate results, and commit identity. Keep human approval on until your evidence supports a narrower auto-merge policy.

Explore the controls →

Before your first run

Questions worth asking.

Can I connect my own repositories?

Yes. Connect authorized GitHub repositories with their own stacks and verification policies. ForgeLoop runs work on your enrolled infrastructure.

Do I need to supply a model API key?

Yes, configure a supported provider on your runner. Provider usage is separate from GitHub authentication. Actual coding work can incur provider charges.

Does connecting a runner start paid work?

No. Browser pairing, saving settings, local key checks, and heartbeat checks do not invoke a model. Start runner can claim eligible work; it is not a dry run.

Is the desktop installer a signed public release?

Not yet. Windows, macOS, and Linux installers are available as unsigned previews through GitHub Releases. Publisher signing and remaining live validation have separate release gates.

Field notes

The engineering behind dependable agent workflows.

Runner operations

Recover a Self-Hosted AI Coding Runner

Use a practical restart recovery checklist for self-hosted AI coding runners: verify durable state, credentials, heartbeats, leases, and safe task resumption.

Read field note

GitHub workflows

From GitHub issues to AI pull requests

Build an issue-to-PR workflow with explicit intake rules, bounded tasks, verification evidence, and a deliberate approval point.

Read field note

Architecture

Coding agents vs. delivery harnesses

Understand the difference between an AI coding agent and the orchestration, policy, evidence, and recovery system around it.

Read field note

Verification

Acceptance criteria coding agents can verify

Write observable acceptance criteria for AI coding tasks, including failure paths, permissions, test evidence, and explicit scope boundaries.

Read field note
Read all field notes →