LLM Observability + Guardrails
Ship safer LLM features with a dashboard your team actually uses.
Sigmoda logs every LLM call, estimates cost, flags risky outputs, and gives you fast filters to debug, triage, and control spend.
Observability and guardrails in one place
Stop hopping between logs, notebooks, and billing exports. Sigmoda keeps cost and quality checks together.
Full-fidelity logging
Capture token usage, cost, latency, and metadata for every LLM call. Prompt/response content capture is optional.
Guardrails that stick
Set max output tokens and banned phrases per project; Sigmoda flags risky responses automatically.
Cost + performance clarity
Calls/day and cost/day charts plus per-model breakdown so you know what to tune first.
Three steps to ship with confidence
Integrate in minutes with a small Python helper and a project key.
Install the tiny Python SDK and add your project key.
Send events from your LLM client; Sigmoda ingests with guardrails applied.
Use the dashboard to slice by route, model, status, or text search and triage flagged events.
Start free, upgrade when ready
Free (Beta): 10k events/mo • 7-day retention • 3 projects. Pro (Early Access): 200k events/mo • 30-day retention • 10 projects.
Free (Beta)
Great for prototypes and early internal usage.
- 10k events/month
- 7-day retention
- 1 user • 3 projects
- Basic dashboards + filtering
- Content capture off by default
Pro (Early Access)
For teams shipping customer-facing LLM features.
- 200k events/month
- 30-day retention
- Up to 10 projects
- Higher ingest rate limits
- Priority support email
Answers for teams shipping AI
Quick details on guardrails, cost tracking, and deployment.
How does Sigmoda flag risky LLM outputs?
Every event is scored for max output tokens and banned phrases per project. Anything that crosses those guardrails is auto-flagged so you can triage before it reaches users.
Can I break down spend by model and route?
Yes. Cost and call charts are grouped by day with model-level breakdowns so you can see which routes use gpt-4o, gpt-4o-mini, or other providers and tune the expensive ones first.
Will this work with my existing stack?
Sigmoda accepts plain HTTPS events, so you can plug in any language. The quickstart shows a Python helper, but the API works from JS/TS, Go, or your favorite SDK.
Is there a free tier?
Yes. Free (Beta) includes 10k events/month with 7-day retention and up to 3 projects. Upgrade to Pro (Early Access) for higher volume, longer retention, and team support.
Do you store prompts and responses?
Sigmoda can store prompt/response content when you enable content capture. In production, content capture is off by default — you can still log metadata and token usage for filtering and cost visibility.