Self-hosted error monitoring, with the AI fix engine included
Run the whole thing in your own boundary: ingest, grouping, alerts and the Conductor fix engine. One Docker Compose stack or a Helm chart, source-available under the FSL, on your own model key. Your errors and stack traces stay in your own databases.
Why self-host your error monitor
Some teams cannot send stack traces to a third party cloud. With Condux they stay in your own databases, and the one thing that can leave is a scrubbed code extract you choose to send to your model provider.
- Your data stays in your boundary
Errors and stack traces are stored in your own databases, and the fix agent runs inside your network. A scrubbed code extract goes to your model provider only when a fix is requested.
- The AI runs on your key
Bring your own Anthropic or OpenAI-compatible key. The spend is on your contract and the model never sees your repo token.
- No per-seat tax
Seats are free. You are not billed per developer to look at your own errors.
- Auditable by design
Source-available code and a full per-run audit trail, so your security team verifies the claims instead of trusting them.
Two ways to run it
Start on one VM, move to a cluster when you need to. Same code either way.
git clone <your Condux fork>cp deploy/.env.example deploy/.envdocker compose -f deploy/docker-compose.prod.yml up -d
helm install condux ./deploy/charts/condux \--set secret.existingSecret=condux-secrets \--set externalStores.clickhouse.url=... \--set config.ingestHost=ingest.example.com
Bring your own managed Postgres, ClickHouse, Redpanda and S3-compatible object storage, or run them in the same stack for a small deployment.
Frequently asked questions
Is self-hosting really free?
Yes. Condux is source-available under the FSL and self-hosting is free forever. The managed cloud is a convenience, not a gate.
Does the AI fix engine run when I self-host?
Yes, and that is the point. The Conductor runs inside your stack on your own Anthropic or OpenAI-compatible key, so the spend is on your contract and the model never sees your repository token. A scrubbed extract of the code does reach the model provider you chose, but only when a fix is requested.
What do I need to run it?
One Docker Compose stack for a single VM, which starts the datastores too, or the Helm chart against your own managed Postgres, ClickHouse, Redpanda and object storage for a cluster. The images are built from the source in the repository, and a migrations job sets up the schema.
Is it Sentry-compatible?
Yes. Point an existing Sentry SDK at your self-hosted Condux by changing the DSN, or report with no SDK at all over native OpenTelemetry logs.