Skip to content

INTEGRATIONS

It connects to what you already run.

One required source of signal, one model provider, and a set of response paths that stay switched off until someone turns them on. Nothing here assumes a particular vendor stack.

HOW DATA MOVES

Three connection points, in one direction.

Signal in

Your error tracker

Issues and event payloads are pulled on a schedule and grouped into one accountable incident.

Reasoning

The AI provider you choose

Redacted context goes to any OpenAI-compatible endpoint — hosted or running on your own hardware.

Response out

The channels you enable

A notification, a pull request, or an approved file change. None of them are on by default.

THE CATALOGUE

Every connection, and whether you need it.

Error tracking

Required

  • Sentry-compatible error tracking

    Point the workspace at your instance, add an organisation, project, and token, and issues start syncing.

    Pull-based · scheduled sync

  • Event payloads

    The full event attached to an issue is fetched for analysis and stored alongside it for review.

    Live fetch when no local copy exists

AI providers

Required

  • Hosted OpenAI-compatible endpoints

    Any provider that speaks the chat-completions API works, including OpenAI and NVIDIA-hosted models.

    Configured in settings · key encrypted at rest

  • Local models

    Run Ollama or LM Studio on your own hardware so incident context never leaves the network.

    Point the workspace at its endpoint · no key needed

Notifications

Optional

  • Email (SMTP)

    Send incident briefs from your own SMTP server to the addresses each project nominates.

    Per-project recipients · cooldown enforced

  • Discord

    Post the summary, cause, and recommended response to a channel webhook.

    Webhook URL per project

  • Telegram

    Deliver the same brief to a chat through your own bot token.

    Bot token · chat ID per project

  • Risk-level routing

    Choose which severities are allowed to reach a person, per project, instead of paging on everything.

    Low · Medium · High · Critical

Change paths

Optional

  • GitHub, GitLab, or Bitbucket

    Open a branch and a pull or merge request carrying the proposed repair, so review happens where your team already reviews. The branch is cut from your default branch and opened against it — never pushed to it.

    One provider per project · scoped to the repositories you pick

  • FTP / SFTP apply and revert

    Apply an approved change to a target and keep the backup taken beforehand, so a revert has something to restore.

    Approval required · backups stored with the analysis

CONNECT WHAT YOU HAVE

Bring your own tracker, your own model, your own channels.

Two connections get you a working incident review. The rest stay off until they earn a place.

  • Any Sentry-compatible error tracker as the source
  • Any OpenAI-compatible model — including a local one
  • Pull requests and file changes stay optional and approval-gated
Talk to us See how the service works

A scoping call, not a sales sequence.