OTLP straight from any SDK, without a Collector
Traceway is an observability platform written in Go. It accepts traces, logs and metrics over OTLP/HTTP straight from any OpenTelemetry SDK, with no Collector in between. Exceptions, session replay, LLM calls and on-call paging share the same dashboard, and the README pitches it against the Prometheus, Loki and Tempo stack you would otherwise wire together yourself.
In server mode, telemetry lands in ClickHouse, the same store SigNoz uses. Superlog also ingests OTLP on top of Postgres and ClickHouse, but hands incident investigation to AI agents.
The open source observability comparison sets Traceway's storage against the other platforms' and looks at the data volume at which each one starts to pay off.
Symbolicated stack traces, LLM costs and on-call paging
- Exceptions symbolicated on arrival: every stack trace gets a SHA-256 fingerprint and joins a ranked issue. The Go symbolicator reads source maps, iOS dSYMs, Android R8 mappings and Flutter obfuscation maps. It also ships as an OpenTelemetry Collector processor, compatible with Honeycomb's
source_map_symbolicator. - Endpoints ranked by impact: P50/P95/P99 latency, throughput and error rate per route, sorted by Apdex.
- Frontend and mobile SDKs: Traceway's own SDKs cover React, Vue, Svelte, Flutter, Android, iOS and React Native. Session replay works on the web and in Flutter.
- LLM call tracking: cost, tokens, latency and full conversations, from OpenRouter or any OTel-compatible AI gateway.
- On-call built in: rotations, escalation policies and pages that keep escalating until someone acknowledges, over email, Slack, Pushover, Telegram or SMS.
- SSO included: Google, GitHub or any OIDC provider, with user roles mapped from an ID token claim.
SQLite and DuckDB images stay single-instance
The deployment mode sets the ceiling. The README's docker compose up -d starts Traceway with ClickHouse 24.8 and PostgreSQL 17. The SQLite image fits in one container, but its docs limit it to trying Traceway out or running a small instance. The DuckDB image is the recommended single-container option, and the same docs say it keeps dashboards responsive with 10 to 100 times more stored rows than SQLite. Both run as a single instance: more than one backend requires PostgreSQL and ClickHouse.
DuckDB memory needs a manual cap. DuckDB budgets roughly 80% of what it believes is host RAM. In a memory-capped container without DUCKDB_MEMORY_LIMIT, the docs warn that the backend gets OOM-killed under ingest.
The repo's compose file mounts no volume for Traceway. Uploaded source maps, session recordings and AI trace payloads are written to /app/storage and are lost when the container is recreated. The docs add that symbolication then silently stops matching. A volume or an S3 bucket prevents the loss.
2.0 was a breaking upgrade. Released on GitHub on September 21, 2026, it changed the trace tables and parts of the API, and the announcement warns of a bigger database and possibly slower reads and writes. History is copied over after a backup by setting V2_MOVE_OVER=true. Four patch releases followed, up to 2.0.4 on October 5. Profiling is still flagged experimental.
The repo is under a year old and has one main developer. Created in December 2025, it published 89 backend releases on GitHub between May and October 2026. One developer wrote 823 of the 1,056 commits on the main branch; the next human contributor has 40.
Nothing is held back for the cloud, according to the project. The README claims MIT with no BSL and no open core, and the website promises full feature parity with Traceway Cloud. The cloud charges for hosting: a free tier with 1 GB of ingest per month, and Premium at $24.99 a month for 150 GB.