Mendry FAQ
Availability: available — individual answers link to pages that carry their own availability labels.
Product
What is Mendry?
Mendry is a self-hosted AI incident investigation system. It connects production alerts, logs, and deployed code, produces an evidence-backed root-cause diagnosis, and can deliver a policy-constrained fix as a review branch or Draft Pull Request/Merge Request.
Is Mendry open source, and what does it cost?
Yes. Mendry is released under the Apache License 2.0 and has no license fee. You pay for your own infrastructure, your LLM provider usage, and any AI host subscription you choose to use.
Do I have to deploy Mendry to try it?
No. Mendry Bot is a single guide that Grok, Meta Muse.ai, or a Claude Code cloud session follows with its own tools to investigate an alert and, when you authorize it, push a fix branch. Self-hosting the full incident application is a separate path for source evaluation.
Try Mendry with your existing bot
Is Mendry production-ready?
Not yet. The incident application runs from source for evaluation, and a supported installation, upgrade, and rollback package is still planned. Each capability is labeled available, preview, or planned according to current evidence.
Safety and control
Does Mendry auto-merge or deploy fixes?
No. Mendry publishes fixes as a review branch or Draft PR/MR. Repository CI and your engineers remain the validation and merge authorities, and merge, deployment, rollback, and recovery stay human-controlled by design.
What can an automatic fix change?
Only what project policy allows: an explicit path allowlist, a denylist that protects credentials, dependency manifests, lockfiles, and runtime version files, and limits of 1 to 30 files and 1 to 5,000 changed lines per attempt. Binary files, symlinks, submodule changes, and diverged target branches fail closed.
Can the model run arbitrary commands or choose credentials?
No. Model output is treated as untrusted proposal data. Service-owned policy validates tool names, scope, budgets, evidence citations, and plans, and the model cannot choose credentials, arbitrary URLs, shell commands, or merge, deployment, or recovery approval.
What data leaves my environment?
Only calls to the LLM provider you configure. Mendry logs safe metadata such as host, path, model, status, duration, and token usage, but does not persist raw prompts, raw model responses, or bearer tokens. Provider API keys are write-only.
Integrations
Which LLM providers work with Mendry?
Any OpenAI-compatible endpoint, through the Chat Completions or Responses API, and Anthropic through the Messages API. The provider is configured per project; if the webhook classifier call fails, Mendry falls back to deterministic normalization.
Which repository platforms are supported for fixes?
Draft PR/MR delivery works with GitHub, GitLab, Gitea, and Yunxiao. Generic Git and Yunxiao publish a review branch directly when the platform cannot create an API-managed change request.
Where can incidents come from?
Signed webhooks from your monitoring system, Tencent CLS callbacks (preview), or the SSH managed log probe, which watches a log file on a Linux host when you have no logging or alerting platform.
Which projects support local pre-validation of fixes?
Single-service Go modules with go.mod and go.sum, and Node/npm projects with package.json, package-lock.json, and a test script. Tests run in a network-disabled container before the Draft PR is opened; unsupported layouts fall back to basic Draft PR mode.
Where can Mendry send notifications?
Telegram, Feishu, and WeCom. Project notifications cover incident triggers and first AI results, with durable delivery and retries.