Traceway

Traceway is an open-source observability platform built on OpenTelemetry, released by dusanstanojeviccs under the tracewayapp organization on GitHub. It brings an app's logs, traces, metrics, exceptions, session replay and on-call alerts into one dashboard. Backends connect by sending OpenTelemetry data to it, and web and mobile apps use Traceway's own SDKs. The AI side has two parts. AI tracing records the cost, tokens, latency and full conversations of LLM calls. The repo also ships two Agent Skills: traceway-setup lets coding agents such as Claude Code, Cursor and Codex read your code and wire up monitoring, and traceway installs the command-line tool (CLI) so the agent can query production errors and logs and follow them back to the root cause. The README says the CLI is read-only apart from archiving exceptions, which needs an explicit --yes. You can self-host it with Docker or use the official Traceway Cloud. What sets it apart is that, according to the README, every feature lives in the same MIT-licensed code, so self-hosting and the cloud run the same thing, with nothing held back for a paid tier.

Who it's for:developers and operations engineers who need to debug production apps or AI agents. Readers who don't write code won't have a use for it.

Before you start: self-hosting needs Docker. The README's docker compose starts Traceway, PostgreSQL and ClickHouse and uses port 80. The database passwords and JWT secret in the config are placeholders; SECURITY.md says they are for local development only and must be replaced in production. Traceway Cloud is an optional paid hosted service; check the official site for pricing. Know what it collects by default before you add it to an app. The web SDK turns on session replay by default: it uses rrweb to record the page (DOM) and the user's clicks, scrolls and typing in 30-second segments, and uploads the recording along with any error. With recordAllSessions turned on, it uploads every segment of every visit. Only password fields are masked by default. Names, emails, amounts, chat content, avatars and the like are recorded unless you add rr-mask, rr-block or rr-ignore classes to your pages yourself, or you can switch replay off with sessionRecording: false. AI tracing stores full conversations and tool calls, and the README suggests setting user.id to a stable customer identifier. All of this goes to the server you host or the cloud you choose, so decide on masking, how you tell your users and how long you keep the data before you connect it. traceway-setup reads and edits your repository: it adds SDKs, writes tokens into env files that aren't committed to Git, and adds source map uploads to CI. Its SKILL.md says it shows you a project plan first and waits for your confirmation before changing code. The CLI install script downloads the latest release to /usr/local/bin or ~/.local/bin and checks the SHA-256 only when the release includes a checksum file. The repo also contains a Skill for the CLI's internal tests and an optional host monitoring agent installer, which AI Inspiration World has not reviewed or recommended. MIT license.

Based on the project's public documentation. Not tested by AI Inspiration World.