Arbor is deterministic PR blast-radius analysis. It parses the repo, builds a call graph, walks changed paths, and comments with what the graph can reach — without an LLM in the risk path.

Can AI coding agents use Arbor?

Yes. Arbor produces a compact PR brief with risky paths, likely side effects, unknown edges, and the first test to write so agents can keep edits scoped.

Does Arbor work with GitHub pull requests?

Yes. Arbor installs as a GitHub App, analyzes pull requests, and comments with the breakage path reviewers should inspect.

Back to Arbor

Changelog

We ship the graph, then we polish the operator loop around it.

This is the public record of what changed in Arbor Cloud: engine work, product shifts, dashboard fixes, launch polish, and the hardening that makes the PR walk usable by real teams.

PR breakage mapdashboard UXagent handoffruntime hardening

0.9.5Major

August 2026

Paid launch, plain-English comments, and a dashboard that fills itself

Arbor left open beta. Pricing is live in dollars, the PR comment now opens in plain English before the numbers, and the dashboard tells you what it is doing instead of showing a blank page. A run of reliability work sits underneath all of it.

Pricing is live: Pro at $9 a month, Team at $60 a month for up to 10 developers, charged in USD.
Every installation gets a free trial of 10 walks. Only completed walks count against it, so a failure on our side never costs you one.
Promo codes for comped access, redeemable from the dashboard.
PR comments now open with a plain-English summary written by a model that is only allowed to rephrase the analysis, never add to it. Every identifier and figure is checked against the computed result before the comment posts.

0.9.0Polish

April 2026

Operator polish and mobile-first launch pass

Arbor now feels like one connected product instead of a stack of promising parts. This release focused on mobile hero quality, dashboard cohesion, auth smoothing, and public pages that finally match the shipped experience.

Rebuilt the landing hero around PR breakage maps instead of a decorative desktop graph.
Improved mobile layouts so the landing page reads clearly on phone-sized screens without losing the product story.
Normalized dashboard lifecycle states so completed analyses stop rendering as in progress across views.
Smoothed first-time GitHub sign-in recovery when a valid session exists after OAuth redirect.

0.8.5Fix

April 2026

Runtime hardening and status quality

Launch reliability work across the API and GitHub integration paths so teams can trust the operator loop around the analysis engine.

Added readiness checks, request IDs, request timeouts, and structured tracing around webhook and worker flows.
Improved commit-status target URLs so failing checks land on the right Arbor run page.
Made status-post failures non-fatal to analysis completion where appropriate.
Improved rate-limit feedback and retry metadata for better operational debugging.

0.8.0Major

April 2026

Breakage analysis pipeline

Breakage analysis is now a first-class stage in Arbor. The engine classifies changed code, walks reachable paths, and emits structured PR notes instead of vague reviewer prose.

Added classifier heuristics across 10 surface categories.
Added upstream and downstream call-path tracing through the graph.
Added framework-aware entrypoint detection for common web stacks.
Added repo rule loading from .arbor/security.yml and .arbor/security.json.

0.7.0Major

March 2026

PR breakage product pivot

Arbor moved from generic blast-radius language to concrete PR breakage paths. The product is now positioned around context for AI-generated pull requests.

Reframed the dashboard around reachable paths, likely side effects, and next checks.
Introduced frontend run contracts to safely consume backend-native path output.
Added controls for monitor, review, block, owner handles, tests, and unknown path handling.
Removed overclaiming language around certainty and universal vulnerability detection.

0.6.0Engine

February 2026

Private engine expansion

The private engine gained multidimensional scoring, weighted signal fusion, and stronger calibration plumbing for richer PR decisions.

Added four-dimensional scoring across correctness, risk, reliability, and maintainability.
Added 21 signals with reliability weighting and freshness decay.
Added conditional verdicts and stronger scenario regression coverage.
Expanded calibration and property-based testing in the private engine path.

0.5.0Major

December 2025

GitHub App and cloud dashboard

Arbor Cloud launched with GitHub installation flow, PR analysis history, and the first production dashboard.

Shipped GitHub OAuth and repository-aware dashboard flows.
Added webhook verification and durable queued analysis processing.
Prepared subscription wiring for a future hosted product tier.
Deployed API on Render and frontend on Vercel.

0.4.0Archive

November 2025

Blast-radius foundations

The original Arbor graph engine shipped with tree-sitter parsing, petgraph topology, and the first generation of upstream and downstream impact analysis.

Built tree-sitter parsing across 14 languages.
Built graph construction with caller and callee traversal.
Shipped the first PR comment with structural dependency impact.

Back to Arbor

Changelog

We ship the graph, then we polish the operator loop around it.

This is the public record of what changed in Arbor Cloud: engine work, product shifts, dashboard fixes, launch polish, and the hardening that makes the PR walk usable by real teams.

PR breakage mapdashboard UXagent handoffruntime hardening

0.9.5Major

August 2026

Paid launch, plain-English comments, and a dashboard that fills itself

Pricing is live: Pro at $9 a month, Team at $60 a month for up to 10 developers, charged in USD.
Every installation gets a free trial of 10 walks. Only completed walks count against it, so a failure on our side never costs you one.
Promo codes for comped access, redeemable from the dashboard.
PR comments now open with a plain-English summary written by a model that is only allowed to rephrase the analysis, never add to it. Every identifier and figure is checked against the computed result before the comment posts.

0.9.0Polish

April 2026

Operator polish and mobile-first launch pass

Rebuilt the landing hero around PR breakage maps instead of a decorative desktop graph.
Improved mobile layouts so the landing page reads clearly on phone-sized screens without losing the product story.
Normalized dashboard lifecycle states so completed analyses stop rendering as in progress across views.
Smoothed first-time GitHub sign-in recovery when a valid session exists after OAuth redirect.

0.8.5Fix

April 2026

Runtime hardening and status quality

Launch reliability work across the API and GitHub integration paths so teams can trust the operator loop around the analysis engine.

Added readiness checks, request IDs, request timeouts, and structured tracing around webhook and worker flows.
Improved commit-status target URLs so failing checks land on the right Arbor run page.
Made status-post failures non-fatal to analysis completion where appropriate.
Improved rate-limit feedback and retry metadata for better operational debugging.

0.8.0Major

April 2026

Breakage analysis pipeline

Breakage analysis is now a first-class stage in Arbor. The engine classifies changed code, walks reachable paths, and emits structured PR notes instead of vague reviewer prose.

Added classifier heuristics across 10 surface categories.
Added upstream and downstream call-path tracing through the graph.
Added framework-aware entrypoint detection for common web stacks.
Added repo rule loading from .arbor/security.yml and .arbor/security.json.

0.7.0Major

March 2026

PR breakage product pivot

Arbor moved from generic blast-radius language to concrete PR breakage paths. The product is now positioned around context for AI-generated pull requests.

Reframed the dashboard around reachable paths, likely side effects, and next checks.
Introduced frontend run contracts to safely consume backend-native path output.
Added controls for monitor, review, block, owner handles, tests, and unknown path handling.
Removed overclaiming language around certainty and universal vulnerability detection.

0.6.0Engine

February 2026

Private engine expansion

The private engine gained multidimensional scoring, weighted signal fusion, and stronger calibration plumbing for richer PR decisions.

Added four-dimensional scoring across correctness, risk, reliability, and maintainability.
Added 21 signals with reliability weighting and freshness decay.
Added conditional verdicts and stronger scenario regression coverage.
Expanded calibration and property-based testing in the private engine path.

0.5.0Major

December 2025

GitHub App and cloud dashboard

Arbor Cloud launched with GitHub installation flow, PR analysis history, and the first production dashboard.

Shipped GitHub OAuth and repository-aware dashboard flows.
Added webhook verification and durable queued analysis processing.
Prepared subscription wiring for a future hosted product tier.
Deployed API on Render and frontend on Vercel.

0.4.0Archive

November 2025

Blast-radius foundations

The original Arbor graph engine shipped with tree-sitter parsing, petgraph topology, and the first generation of upstream and downstream impact analysis.

Built tree-sitter parsing across 14 languages.
Built graph construction with caller and callee traversal.
Shipped the first PR comment with structural dependency impact.