Agent eval · observability lite
Know what your agents actually did.
EvalLite records runs, trace spans and scores through one small API. Prompts stay redacted unless you opt in. Quotas are counted in plain events.
Free plan · 50,000 events/month · no card
curl https://evallite.com/v1/runs \
-H "Authorization: Bearer evl_••••" \
-d '{"name":"checkout-agent@v42"}'
→ 201 { "id": "run_9f…", "status": "open" }
curl https://evallite.com/v1/runs/run_9f…/traces \
-d '{"spans":[{"name":"plan","prompt":"…","output":"…"}]}'
→ 201 { "redacted": true,
"spans": [{ "prompt": "[REDACTED]" }] }How it works
Three objects. That's the whole model.
01
Run
One execution of an agent or eval suite. Tag it with a name and metadata.
02
Trace
Spans of what happened — tool calls, plans, outputs. Prompts redacted by default.
03
Score
A metric and value with a short reason. Compare runs over time.
Redacted by default
Prompt and output become [REDACTED] unless you send store_raw: true.
Plain event quotas
One run, span or score is one event. Records over 64 KB are rejected, not billed.
Hashed API keys
evl_ keys are shown once at creation. Only a SHA-256 hash is kept.
Retention you can read
Data older than your plan window — 14 to 365 days — is purged daily.
Plans
Pay for events, not seats.
Free
Free
50,000 events / month
For prototypes and side projects. Hard cap — writes stop at the limit.
- 14-day retention
- 429 at the monthly cap
- Redaction on by default
Starter
US$49/ month
US$490 / year · 250,000 events / month
For a single production agent. Keeps accepting past the cap.
- 30-day retention
- Overage recorded, never blocked
- Redaction on by default
Team
PopularUS$199/ month
US$1,990 / year · 2,500,000 events / month
For teams running eval suites in CI on every change.
- 90-day retention
- Overage recorded, never blocked
- Redaction on by default
Scale
US$799/ month
US$7,990 / year · 15,000,000 events / month
For fleets of agents and year-long audit trails.
- 365-day retention
- Overage recorded, never blocked
- Redaction on by default