endr Connect GitHub →

Continuous AI dependency monitoring · no API key

Stop AI model retirements
before they break production.

Connect GitHub once. Mendr continuously monitors your AI dependencies, explains the risk, and prepares a validated migration pull request for human approval.

Connect GitHub Run it locally

See it work

How it searches your code, finds the problem, and fixes it.

The audit walks every file, tells a live call from catalog data, and matches the registry. fix-llm goes one step further: it writes the exact diff and proves it before anything is applied. Connected through the App, you approve the migration on the finding and your own CI opens the pull request. You watch each step and see the change itself without leaving Mendr.

mendr fix-llm · services/geminiService.ts 0:18
scanfindclassifymatchchangeverifydecide
$ mendr fix-llm . walking 342 files · typescript + python 3 references to gemini-2.0-flash services/geminiService.ts:18 call argument config/pricing.ts:22 object key · catalog tests/mock.spec.ts:9 test fixture classify · services/geminiService.ts:18 ai.getGenerativeModel({ model: MODEL_NAME }) → live call catalog + fixture are data · left alone registry · gemini-2.0-flash retired replacement gemini-flash-latest · listed in public catalogs diff · services/geminiService.ts:18 - const MODEL_NAME = "gemini-2.0-flash" + const MODEL_NAME = "gemini-flash-latest" gate · patched copy ✓ type-check ✓ tests (your own suite) Tier A · verified · 1 change ready to review, catalog untouched

A fix-llm run on an example repository. Press play to watch it step by step.

Works with

The LLMs your product already calls.

Deprecated model ids and the params coupled to them, tracked across every major provider, in TypeScript, JavaScript and Python.

Every record links to the provider's own documentation. A weekly job checks that each verified replacement is still listed in public model catalogs.

Detect

Know what is retiring before the notice lands.

Mendr keeps a sourced deprecation registry and flags the models your systems depend on, ordered by deadline.

  • Sourced from the provider's own documentation
  • Ordered by nearest deadline
  • OpenAI, Anthropic and Google

Optional · Measure

Prove which ones are live. If you want to.

Code says where a model is declared, not whether production calls it. Connect telemetry you already have and the same finding gains an observed-usage line. Until then the report says not measured, never "unused".

  • OpenTelemetry, a usage export, your own key in your own CI, or gateway logs
  • Reads provider, model, service, timestamp and volume. Never prompts or responses
  • Never required. Absence is never treated as proof

Locate

Find the configuration that selects it.

Mendr separates a live selector from catalog data, so your team changes the right line, and only the right line.

  • Config and IaC: yaml, json, .env, helm
  • Selector vs. catalog, told apart
  • Reported as a candidate for review
DECISION / WITHHELD
CANDIDATE SELECTOR
app.yaml:3
EVIDENCE MISSING
Runtime reader tie-back
REQUIRED
Human confirmation
APPLIED
No code changed

Evidence

One route, read as evidence.

Each signal carries its evidence, its source, and how far it can be trusted. Nothing is asserted past what the data shows.

SignalEvidenceSourceConfidence
Retirement2026-10-23RegistryVerified
Usagenot measuredRuntime · optionalUnknown
Locationsrc/ai/client.ts:42Source scanVerified
DecisionPatch eligibleSafety gateNot applied
RETIREMENT WINDOW
ANNOUNCED
provider notice
NOW
mendr audit .
DEADLINE · gpt-4
2026-10-23
RESULT

Call site located and verified. Migration evidence attached. Runtime usage not measured, and reported that way.

The commands

A small set, run in your own environment.

mendr audit [path] · previewThe audit. Source, config and registry in one report with four honest conclusions. Needs only the repository.
mendr audit . --install · previewScaffolds a GitHub workflow that keeps one issue per repo current: new, continuing, resolved. contents: read, issues: write.
mendr fix-llm [path]Print the gated diff for a retired id at a verified call site, proved against your type-check and tests before it touches your code. Released, v0.1.0.
mendr watch [path]The lightweight tracker: which deprecated ids a repo uses, by deadline, in one self-updating issue. Released, v0.1.0.

Install

Run the audit. Under a minute, no key, nothing changed.

Point it at a repo. It scans TypeScript, TSX, JavaScript, Python and config, joins the retirement registry, and prints every retiring model with its location, deadline and migration evidence. Add --install to keep one GitHub issue current.

audit · v0.5.11-alpha
$ npx github:ajitheee/mendr#v0.5.11-alpha audit . mendr audit (preview) ✓ Source code 342 files (301 TS/TSX, 41 Python) ○ Configuration not applicable ✓ Registry anthropic, google, openai ○ Runtime usage not measured (optional) Conclusion: EXPOSURE DETECTED gpt-4 src/ai/client.ts:42 2026-10-23 verified direct call site · replacement gpt-5.6-sol Decision: PATCH ELIGIBLE · No change applied

Pricing

Free while it is in alpha. Free for a year for five design partners.

Everything mendr does today is free. Before setting a public price, we are working with five design partners whose product breaks when a model id goes dark. They help shape mendr, and use it free for 12 months.

Free available now

$0

  • The CLI, unlimited, in your own environment: audit, fix-llm, watch
  • The retirement registry for OpenAI, Anthropic and Google
  • The GitHub App: the deadline on the check run, one self-updating issue per repo
  • Fix pull requests from your own CI, only after the swap passes your type-check; a failing test suite blocks it
  • Our servers never clone or read your repository
Run the audit

Design partner 5 teams

$0 for 12 months

  • Everything in Free
  • Set up with you on your main product repository, in your own CI
  • We work through each retirement that hits your code with you, starting with OpenAI's 2026-10-23 wave
  • A direct line to the person who builds mendr
  • In return: your feedback on what breaks, what is missing and what you would pay for
  • After 12 months, the regular price of the plan you choose
Apply to be a design partner

Regular plans planned, not on sale

Team $49/mo or $490/yr, 2 months free

Up to 10 private repositories watched. Next-business-day support.

Business $199/mo or $1,990/yr

Unlimited repositories. Same-day support from a named contact.

Priced per company, never per seat: the risk grows with the model calls in your code, not with the size of your team.

Prices in USD. Planned prices may change before they go on sale. Design partners move to the regular price after their 12 free months.

Get started

Know what retires under your code before it becomes an incident.

Connect GitHub Run it locally Book a read-only audit