Local-first · Works with your AI coding agent

Ship AI code fast. Without shipping the mistakes.

Review branches before you push, understand large PRs through guided walkthroughs, and track findings with your own AI provider. You decide what to fix, approve, and publish.

or email contact@loomtide.ai

Free and open source · Works with Claude Code, Codex, Gemini, and OpenCode

Branch review

Review every commit, not just every PR.

Run a branch review before you push or open a pull request. Use a supported AI provider to inspect the change and its context from your local workspace.

  • Optional review on commit

    Enable automatic review for a project to review new commits while Atlas is running.

  • No PR required

    Review happens locally, before you push to GitHub.

  • Fix while context is fresh

    Review the branch while the change is still fresh, then inspect the findings and decide what to fix.

PR review

When the PR opens, Atlas keeps going.

Atlas isn't only a pre-PR check. The same review engine handles your full PR loop - without leaving the desktop, without paying for a second tool.

PRs organized as Kanban

Keep ready-to-review PRs, your own changes, and completed work organized in one Kanban inbox.

Step-by-step walkthroughs

Follow logical steps through a change, with explanations and diffs side by side. Large PRs stay responsive with virtualized file rendering.

Ask Atlas Chat

Hover any finding or diff hunk and ask the review engine for context - what changed, what the impact is, how to test it.

How it works

From a code change to an informed review.

Review locally, use your own AI provider, and keep the decision in your hands.

  1. 01

    Agent commit

    Trigger

    Start a branch review manually, or enable automatic review for a project to review new commits while Atlas is running. No PR is required.

  2. 02

    Policy checks

    Deterministic

    Use project review settings and .atlas/review-policy.json to guide the review. Configure supported checks and context for your codebase; available checks depend on your tools and setup.

  3. 03

    LLM review

    Bring your provider

    Atlas sends selected code and context to your configured AI provider and returns structured findings with file paths, severity, and suggested fixes. Your provider charges and usage limits apply.

  4. 04

    Review decisions

    Your decision

    Inspect findings, track them as addressed or dismissed, and choose what to publish to GitHub. Atlas does not automatically configure required GitHub checks or branch protection.

  5. 05

    Follow through

    Track outcomes

    Keep review history and finding outcomes in the same workspace. Your team decides when to merge and configures repository protections separately.

Local-first

Your workspace stays under your control. Choose the services you use.

Atlas Review runs locally and stores review state on your disk. Reviews and chat send selected code and repository context to your configured AI provider. Use a provider and account appropriate for the code you are reviewing.

YOUR MACHINErepoAtlasreviewfindings · settings · historystored locallycode +contextYOUR LLM PROVIDERClaude CodeCodexGeminiOpenCodeyour provider account
Selected code and context go to your AI provider. GitHub features contact GitHub. Optional hosted sync can also transfer review data.

Read the privacy and data flow details

  • Local review workspace

    Project state, findings, and review history are stored in your local profile. Local review does not require an Atlas-hosted service.

  • Your provider, your account

    Your AI provider receives the selected code and context needed for review. Provider CLIs can read additional repository files. Your provider terms and usage limits apply.

  • Optional hosted sync

    Fresh installs have hosted sync disabled. Existing saved connections are preserved on upgrades and can transfer review state, findings, and project metadata to the configured service.

Policy as code

Your repo. Your rules. No dashboard.

Atlas reads review policy from a file in your repo. Edit it like any other code, review it in pull requests, and roll it back if you do not like the change.

  • Versioned with your code

    Config lives in .atlas/review-policy.json, tracked in git like everything else.

  • Reviewed in PRs

    Policy changes ship the same way feature changes do. Your team approves them.

  • Per-path control

    Set review depth and add path-specific guidance for backend/, frontend/, docs/, or anything else.

.atlas/review-policy.json
{
  "version": 1,
  "defaults": {
    "requireEvidence": true,
    "maxRecommendations": 8
  },
  "linters": {
    "enabled": true,
    "mode": "project_only",
    "eslint": {
      "enabled": true
    }
  },
  "paths": [
    {
      "glob": "backend/**",
      "fileClass": "deep",
      "promptAppend": "Check authorization and error handling."
    },
    {
      "glob": "docs/**",
      "fileClass": "shallow"
    }
  ],
  "promptAppend": [
    "Prioritize correctness and security over style."
  ]
}

Signal, not noise

Know if your AI reviews are useful.

Atlas tracks which findings you address, dismiss, or leave pending. Use that history to understand which review recommendations are useful and where to adjust your workflow.

  • Every finding has an outcome

    Addressed, dismissed, or still open. Atlas keeps the record so nothing falls through the cracks.

  • Measure dismiss rate per model

    If gpt-5 is producing 90% useful findings and another model is at 40%, you'll know - and you can switch.

  • Tune the policy with evidence

    Path-level dismiss rates show where review is too noisy or too loose. Adjust the config, not the gut feel.

LAST 30 DAYS · ProjectBlack
138findings
79addressed
11dismissed
48pending

Useful signal·only 8% dismissed

Get Atlas Review.

Choose the build for your computer. On Mac, check About This Mac: Chip means Apple Silicon; Processor means Intel.

Apple Silicon is updated to 0.25.27. Intel Mac and Windows retain the 0.25.26 builds and do not include the newer fixes. Install the Apple Silicon update manually; the shared automatic-update feed remains on 0.25.26. Windows installers are not Authenticode signed and may show a SmartScreen prompt.

Atlas Review is free under the MIT license. Your AI provider's charges and usage limits apply. Linux installers are not currently published. Release notes and checksums