Question 1

What is AgentLoom?

Accepted Answer

AgentLoom is the orchestration layer for AI software delivery: it orchestrates Claude-powered coding agents into governed delivery teams that run the software delivery lifecycle — plan, code, CI, blocking review, merge, repair, and follow-up — end-to-end through GitHub. The team lives inside your repo and stays to keep main green after merge — see /product for the full mechanism.

Question 2

How is this different from a coding assistant?

Accepted Answer

A coding assistant is a tool a developer drives — you supervise, steer, and re-prompt as session context drifts. AgentLoom replaces that loop with orchestration: different agent teams run different phases of the lifecycle — plan, code, CI, blocking review, merge, follow-up, post-merge repair — composed and recomposed per task and per phase.

Question 3

How is this different from Devin or GitHub Copilot coding agent?

Accepted Answer

Single-agent systems are bounded by session context and by how many agents a developer can operate in parallel — a human still drives the loop. AgentLoom removes the operator from the loop, not just the typing: it orchestrates many bounded sessions across parallel ephemeral environments, so the loop runs for days with no intervention. When AgentLoom hits a blocker that needs a human, it surfaces it. Single agents run and stop; AgentLoom keeps running until the work is done.

Question 4

Is AgentLoom an agent harness?

Accepted Answer

Yes — in the emerging sense of the term. Harness engineering is the design of the environment and scaffolding around a coding agent: tools, permissions, verification loops, memory, observability. AgentLoom is that harness delivered as a governed product, plus the orchestration layer that runs it across the whole delivery lifecycle. See /product for the mechanism.

Question 5

Do I need to bring a roadmap to start?

Accepted Answer

No. AgentLoom can start from an idea, an existing repo, or an existing backlog. Onboarding runs deep repo exploration — your tech stack, what's implemented, what's tested, what's claimed in docs but not backed by code, and where the design gaps are. From those findings, AgentLoom builds a roadmap aligned to your goals and composes an agent team tailored to your stack and the work ahead — and all of it keeps adapting as the project grows.

Question 6

What languages and frameworks does AgentLoom support?

Accepted Answer

No fixed support matrix. The agent team is shaped to whatever your codebase actually uses, so support follows your stack rather than a published list. If your stack is unusual, talk to sales.

Question 7

Can AgentLoom really run for days?

Accepted Answer

Yes. Single coding agents are bounded by context, scope, and time per session. AgentLoom decomposes the lifecycle into many bounded sessions, runs them in parallel ephemeral environments, and carries state across the run. The result is continuous execution that can clear whole backlogs without needing to restart the system at every step.

Question 8

Does the agent need human supervision?

Accepted Answer

Not continuously. AgentLoom is built for autonomy with the gates you define: it waits for sign-off where you require it, such as before merge, before connector writeback, or before specific state transitions, and keeps moving otherwise.

Question 9

If AgentLoom runs the loop, what do I do?

Accepted Answer

You steer. The delivery loop — backlog to merged PR to follow-up — runs itself. Direction stays yours: feed new feature requirements, bug reports, and priority calls through the dashboard chat, and they become scoped roadmap work, then backlog, then shipped code. And you hold the gates: approvals, merge policy, and connector writeback stay wherever you put them.

Question 10

What happens if a run fails?

Accepted Answer

AgentLoom self-recovers from most run-time issues: red CI gets failed jobs retried, or the team re-runs with the failure context to repair the branch; merge conflicts get auto-resolved when there's no real conflict, or routed back to the team with the conflicting files in context; post-merge breakage on main gets a repair PR within minutes. Anything that genuinely needs a human — a missing credential, an external dependency, a design call — surfaces as a structured blocker, pauses the run, and resumes when you unblock it.

Question 11

How are agent teams composed?

Accepted Answer

AgentLoom analyzes your codebase, your stack, and your roadmap, then assembles role-specialized agents shaped to the work ahead. Each run is composed fresh — tools, access, context, and success criteria scoped to the task at hand. No bloated skill stacks; no generic do-anything agent.

Question 12

What is a follow-up task?

Accepted Answer

When work surfaces adjacent bugs, missing tests, or unanswered design questions, AgentLoom turns them into reviewable follow-up items in your backlog. Backlog quality grows with delivery, not against it.

Question 13

How does AgentLoom access my GitHub?

Accepted Answer

Through a scoped GitHub App installation. AgentLoom uses per-repo installation tokens for the repository being worked. It does not hold standing org-wide credentials or store user passwords. See /security for the access model.

Question 14

Will AgentLoom merge automatically?

Accepted Answer

Only if you turn auto-merge on for that repo. Even then, merge only happens after every assigned review role has approved and CI is green — usually after several review-and-fix cycles. Auto-merge is a "merge once the team has signed off" gate, not a bypass. Review roles come from your org chart, so they match how your team actually reviews. Default is human review and merge.

Question 15

Do agent commits look like real commits?

Accepted Answer

Yes — agent commits land in real GitHub PRs with role-tagged commit messages. Reviews are blocking until every role approves; CI runs as it does for any PR. The workflow is real GitHub work.

Question 16

What if AgentLoom breaks main after merge?

Accepted Answer

An agent team opens a repair PR within minutes. Post-merge integrator behavior is part of the lifecycle, not an add-on. CD-side workflow failures (such as Deploy-staging breakage) are also covered.

Question 17

Does it work with monorepos and polyrepos?

Accepted Answer

Both. Monorepos: agents are scoped per task to the relevant area. Polyrepos: each connected repo gets its own setup. AgentLoom analyzes the repo at setup and adapts.

Question 18

Why delivery capacity instead of per-seat pricing?

Accepted Answer

AgentLoom's cost is driven by execution, not by how many teammates can see the workspace. Seats are free; plans scale with delivery capacity. See /pricing for the detailed model.

Question 19

Does my subscription cover normal delivery?

Accepted Answer

Yes. Each plan caps the agent-work spend AgentLoom will orchestrate in a billing period — totaled from the usage your Anthropic key reports. Runs pause when you reach the cap.

Question 20

How does billing work?

Accepted Answer

Two charges, separately. Anthropic bills you directly for model usage on your own API key. AgentLoom charges a flat monthly fee for the orchestration — the layer that keeps the system running continuously for days: resolving backlog items, handling review and CI, repairing post-merge regressions, triaging follow-ups back into the queue, generating new issues from your roadmap, and refreshing the roadmap as work lands. Each plan caps the agent-work spend AgentLoom orchestrates — totaled from what your Anthropic key reports. Runs pause when you hit the cap; upgrade or wait for the next period to resume.

Question 21

Is there a free tier?

Accepted Answer

Yes. Free includes a real roadmap-planning run plus $25 of orchestrated agent work per month, on one repo. Bring your Anthropic key; no credit card required.

Question 22

Can I switch plans, cancel, or get a refund?

Accepted Answer

Plan changes take effect at the next billing cycle. Cancellation stops future charges. In-flight runs complete; no new ones start. See /pricing for full billing FAQs.

Question 23

Is AgentLoom SOC 2 compliant?

Accepted Answer

AgentLoom is in active development of its formal compliance program. We do not currently claim SOC 2, ISO 27001, or other third-party certifications. The controls in /security are designed to support customer evidence collection, vendor review, internal audit, and incident review.

Question 24

Does AgentLoom support SSO?

Accepted Answer

Enterprise SSO via SAML is supported. SCIM provisioning and granular RBAC are planned Enterprise capabilities. Email and GitHub OAuth remain available for self-serve teams. See /enterprise for the identity model.

Question 25

How does AgentLoom handle our code?

Accepted Answer

Code is fetched per run from your GitHub installation, processed in an isolated ephemeral environment, and discarded after the run. AgentLoom does not maintain long-term mirrors of your repositories. See /security for the data-handling detail.

Question 26

Can I revoke AgentLoom's access?

Accepted Answer

Yes. Uninstall the GitHub App from your org and AgentLoom loses access immediately. Approval gates and connector revocation are also configurable from the dashboard.

Question 27

What about audit logs?

Accepted Answer

Run lifecycle events, status transitions, approval decisions, credential access events, commits, review outcomes, and artifact records are logged with timestamps and event cause. Designed for evidence collection, internal audit, and incident review.

Question 28

How do I sign up?

Accepted Answer

Get started at /signup. Free is no-credit-card and includes a real roadmap-planning run plus $25 of orchestrated agent work per month.

Question 29

Can I invite my team?

Accepted Answer

Yes — seats are free on every plan. Add as many teammates as you need. AgentLoom does not charge per seat.

Question 30

How do I get help?

Accepted Answer

Free: community and docs. Pro: community plus email support. Max: priority email. Enterprise: dedicated. For active workspace issues, contact us through /contact.

Question 31

Is there a public API or CLI?

Accepted Answer

AgentLoom's primary surface is GitHub plus a dashboard. Talk to sales about API and CLI access available today and on the roadmap.