What's New

DEMO Explore Torrix without signing up. All data is read-only. Deploy free instance →

Your API Key

Use this in the Authorization: Bearer ... header when proxying requests.

API Key

Loading…

How to use

Send any HTTP request through Torrix by pointing it at the proxy:

curl -X GET http://localhost:8088/proxy \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "x-target-url: https://api.openai.com/v1/..." \
  -H "x-upstream-authorization: Bearer sk-..." \
  -H "x-torrix-name: my-run"

Edition

Plan
Loading…
Data retention
Max runs shown
Golden runs retention
90 days
Max golden runs
Your data is wiped after 7 days. Upgrade to Pro for 30-day retention, unlimited runs, and full team access. Get Pro →

Pro License

Paste your license key to activate Pro features on this instance.

Don't have a key? Get Torrix Pro →

Keep Torrix up to date

Pull the latest image and restart. Your data is preserved.

docker compose pull
docker compose up -d

Product Tour

Restart the onboarding walkthrough to discover features.

API Keys

Manage API keys for SDK integrations, CI pipelines, and automation. Each key can be scoped to specific projects with read-only or full access.

Loading...

Budget Controls

Set a soft alert threshold and a hard cap. The hard cap blocks all proxy requests once your daily spend is reached. No more calls go through until tomorrow.

When the hard cap is set, requests to /proxy return 429 once the daily spend exceeds this amount.

Torrix POSTs {"event":"budget_exceeded","period":"daily","spent_usd":...,"threshold_usd":...}

Slack webhook URLs (hooks.slack.com) are automatically formatted as native Slack messages.

Cost anomaly detection
Enable spike detection
Badge and alert when a run costs more than Nx the model's p95 baseline. Requires at least 10 prior runs for the model.
x the model's p95 cost triggers an alert

Per-project and Per-key Limits

Set daily caps for individual projects or API keys. Requests that exceed a scope cap return 429 until midnight UTC, even if the global hard cap has not been reached.

Request Sampling

Set an instance-wide default sample rate. Per-request x-torrix-sample header overrides this. Errors are always logged regardless of sample rate.

Judge LLM

Default provider, base URL, and model for auto-scoring runs. The API key is stored server-side and used for online evals.

Online Evals

PRO

Automatically score every new production run via the LLM judge as it arrives. Requires a judge API key saved above.

Model Routing

PRO

Auto-rewrite model names on proxy requests. When a matching rule fires, the model field is swapped before forwarding upstream.

Custom Model Pricing

PRO

Set per-token pricing for fine-tuned or private models that are not in Torrix's built-in price list. Cost tracking will use these rates for matching model names.

Audit Log

Security-relevant actions across all users.

Loading...

PII Masking

Detected patterns are replaced with labeled tokens before being stored. Applies to new runs only. Existing data is not modified.