Skip to content
Memoturn

open source

Memoturn

Trace every LLM call, track cost and latency, run offline, online, and human evals, and ship versioned prompts with deployment channels — on infrastructure you own.

Get startedSee use casesView on GitHub
run it

One command, full platform

Memoturn is an open-source AI engineering platform: traces with a waterfall timeline, cost/token/latency metrics over Apache Doris, offline, online, and human evaluation, a versioned prompt registry, a multi-provider playground, and datasets — all behind one console you self-host.

bun run setup && bun run dev

Built on an async ingest pipeline — the API acks your batch, a worker maps it into Apache Doris — with Postgres for metadata, blob storage for the replayable raw event log, and Redis for the queue. Ingest from the TypeScript, Python, or Go SDK, the OpenAI and LangChain wrappers, or any OpenTelemetry exporter. Read the architecture overview for the full data path.

observeOpenTelemetry-native

OTLP/JSON with GenAI semantic conventions — no proprietary agent required.

evaluateEvals, three ways

Offline experiments, online evaluators over live traffic, and human review queues.

connectMCP server

Prompts, datasets, and review queues as tools inside Claude Code, Cursor, and other agent IDEs.

open source · Apache-2.0

See every LLM call, on infrastructure you own.

Traces, evals, metrics, and prompt management — no proprietary cloud required, and no telemetry leaving your network. Run it locally with bun run setup, or deploy with docker compose. See Deployment for the path.

Get startedGet the source