Per-org pricing, bring your own AI key. Your team approves, edits, or drops every finding before it hits GitHub. Efficient by design — we win when your token bill stays small.
Free tier — 5 trial reviews on Haiku · 1 repo · no credit card.
What you get
Manager control, codebase context, per-repo rules — wired through the entire pipeline, not bolted on after the fact.
Pending queue + per-finding curation. Edit, re-sever, or drop AI findings before they hit GitHub. Or auto-post when you trust the bot enough.
Vector retrieval, import graph, reference resolution, and a search-code tool. The reviewer reads beyond the diff before it flags anything.
Plug in your Anthropic or Groq key. Encrypted at rest. You pay the token bill direct — no managed-quota surprises, no markup.
Opt-in on BYO: trivial diffs auto-downgrade to a cheaper model on your own provider — Claude → Haiku, Groq → Llama 3.1 8B. Everything else runs full-strength. The savings are yours, not ours.
Drop a .mergelint.yml in your repo: disable finding types, set severity floors, ignore paths, inject custom guidance into every review prompt.
👍/👎 on review comments — directly on GitHub or in the dashboard — train per-repo confidence. Noisy finding types fade out over time.
A failed GitHub Actions run triggers a second review pass with the CI logs as context. Catches what the diff-only pass missed.
How it works
Mergelint sits between GitHub and your team. Install the App, open a PR, get reviews you trust — with the dial between auto-post and manager-approval set however your team likes it.
One-click GitHub App install. Pick which repos to connect — public or private, your account or your org.
On install, Mergelint embeds your repo so reviews understand the codebase beyond just the diff. Re-indexed on every push.
Mergelint reads the diff plus relevant codebase context, then posts severity-ranked inline comments. Tool-use loop verifies findings before flagging.
Per-repo mode: auto-post findings, or queue them for manager approval. Edit severity, drop noise, sign off — then the bot posts a curated review.
Pricing
Per-org pricing, not per-user. AI tokens go straight from your account to Anthropic or Groq — no markup, no surprise bills. We promise to use them efficiently (see the FAQ for how).
Try the full product on 5 PRs. No API key required.
Includes
Bring your own AI key. Unlimited reviews.
Includes
For teams shipping daily. Per-org, not per-user.
Includes
Your code stays yours. We never sell it, share it, or train models on it. The only third party that sees your diff is the AI provider you chose. Ask us to delete your org and we wipe your embedded code and review history for good.
Need a higher limit? Full pricing & quotas →
FAQ
The honest answers — not the ones the website usually gives you.
Not today — Mergelint is a hosted SaaS. The review engine, codebase indexer, and dashboard live in private repos. If self-hosting matters for your team, get in touch (info@makama.dev) and we can talk about what's feasible.
Your code is sent only to the AI provider you choose (Anthropic, Groq, or your own key in BYO mode) and only to generate the review. We don't store full file contents long-term — only the chunks needed for the codebase index, scoped to your org. We never use your code to train any model, and BYO mode means the platform doesn't store your provider key in plaintext (AES-256-GCM encrypted at rest).
On paid plans — required. Bring your own Anthropic or Groq key in Settings; you get unlimited reviews and pay the token bill direct to the provider, no markup. The key is encrypted at rest with a per-record salt and versioned cipher (v1:), so we can rotate our platform encryption secret without re-encrypting yours. Free tier uses a small managed-mode trial so you can evaluate without an API key first.
Aligned incentives. If we mark up your AI tokens, we have a quiet incentive to spam more findings (more tokens = more revenue). With BYO you pay Anthropic direct and we never see a cent of token revenue — so we're aligned with you on keeping the bill small. That's why every paid plan uses BYO. Concretely we won't bloat tokens because: (1) trivial diffs skip the LLM entirely, (2) the tool-use loop verifies findings before flagging, (3) the verifier model re-checks each finding on a cheaper pass, (4) the memory loop suppresses finding types you've downvoted, (5) repo rules disable categories you don't want, (6) generated files never reach the LLM, (7) incremental indexing only re-embeds changed files, (8) cached repo rules avoid redundant work between back-to-back reviews.
An opt-in setting for BYO orgs (Settings → AI Provider, off by default) that downgrades the reviewer to a cheaper model on your own provider — Claude → Haiku, Groq → Llama 3.1 8B Instant — on diffs our heuristic judges genuinely trivial: small, low-risk, no new imports or function/class definitions, and never on auth/security/migration/routing paths. Everything else keeps running on the model you chose. Since you pay your own provider bill on BYO, every downgraded review is money back in your pocket, not ours — same aligned-incentives logic as the rest of the BYO design. A repo can opt out entirely via `.mergelint.yml: always_use_selected_model: true`.
Deep, codebase-aware reviews — pulling in related files, import graph, and retrieved context — currently cover JavaScript, TypeScript, and Python. Other languages still get a diff-level review (severity-ranked findings on the patch), just without the repo-context features. Broader coverage is on the roadmap.
Three layers: a tool-use loop where the reviewer can lookup_symbol / read_file / search_code to verify a finding before flagging it; a verifier pass on a separate model that re-checks each finding; and a memory loop where 👍/👎 feedback trains per-repo suppression — finding types your team consistently downvotes fade out. For teams that want stronger control, manager-controlled mode queues every review for human approval before posting.
Yes. The GitHub App authenticates via installation tokens — Mergelint only sees repos you explicitly grant access to during install. You can revoke access at any time from your GitHub settings, which disables reviews on the affected repos immediately.
Owner-only one-click via the LemonSqueezy customer portal in Settings → Billing. Cancellation takes effect at the end of the current billing period; you keep paid-tier access until then, after which you drop to Free (reviews keep working, just within Free quotas).
Not yet. Mergelint is single-tenant SaaS today. If you have compliance requirements that need on-prem or VPC-isolated deploy, reach out at info@makama.dev and we can talk about timeline.
Still have questions? info@makama.dev.
Paste a public GitHub PR or a raw diff.
Private repo? Install our GitHub App for reviews on every PR. Sign up or sign in.