Torrix
GitHubStart free

Torrix Observability · Free and source available

Control every token your company spends.

Real-time visibility into AI costs, traces and failures. Self-hosted in your infrastructure. Your data stays yours.

Try the live demo

Works with

Problem

AI spend arrives as an invoice. Failures arrive as support tickets.

The invoice

By the time the bill is in, the month is over.

The support ticket

Neither tells you which request, which agent or which prompt caused it.

Insight

Every request. Every trace. Every cost.

Torrix sits in the path of every call you route through it. Point a client at the proxy, add an SDK, or send OpenTelemetry. Nothing to rewrite.

  • Agents
  • Apps and SDKs
  • Claude Code
  • OpenTelemetry
trace · tokens · cost
,
requests

Live demo

See what your AI is actually doing.

Pick a trace, then a step. Switch to Costs for spend by model, budgets and hard caps.

Runs / Refund request, order 48213

Traces

Calls3
Tool calls2
Subagents1
Tokens13,120
Cost$0.0290
Wall clock6.4 s
StepTimelineTime
Building the agent tree

Pick a step.

Live

Architecture

Your models. Your infrastructure. Your data.

  • Runtime

    One process and one SQLite file. No queue or cluster to run.

  • Scale

    Postgres, Docker, Helm or a single binary when you need them.

  • Privacy

    Data stays on your servers. PII masking covers proxy, SDK and OpenTelemetry traffic.

  • Migration

    Helicone-compatible headers, so switching is a base URL change.

  • Clients

    HTTP proxy, Python and TypeScript SDKs, OpenTelemetry, and .NET and Java clients in the repository.

Start

Start in a minute.

Run one Docker command and open localhost:8088. Or leave your email and we will walk you through a setup for your stack.

Docker
$ docker run -d --name torrix \
    -p 8088:8088 \
    -v torrix_data:/data \
    gettorrixai/torrix:latest
Open http://localhost:8088, create your account, and go to Integrations.
Compare editions