Pricing

Pay for evaluations, not traces.

Iris scores every agent output for quality, safety, and cost. The evaluation is the value unit — that's what you pay for. Traces are commodity. Start free.

Free

For solo devs + open source projects

$0 / month

Install
  • 10,000 evaluations / month
  • All 12 built-in eval rules
  • Custom Zod rules (unlimited)
  • Dashboard + playground
  • stdio + HTTP transports
  • Community support (GitHub Issues + Discord)

Stays free for personal projects. No credit card required.

Pro

For small teams in production

$25 / month base + usage

typical team $5–$10K / year

Join waitlist
  • 100,000 evaluations / month included
  • $0.0005 per evaluation above that
  • All Free-tier features
  • LLM-as-judge rules (v0.4)
  • Trace comparison (side-by-side)
  • Cost breakdown by agent + rule
  • Agent-level dashboard filtering
  • Email support, 48h SLA
  • Self-Calibrating Eval beta access (v0.5)

Billed monthly. Cancel any time.

Enterprise

For organizations with volume + compliance needs

Custom

Contact sales
  • Usage-based pricing, committed volume discount
  • All Pro-tier features
  • Single sign-on (SAML + OIDC)
  • Priority support, 4h SLA on P1
  • Custom eval rule authoring services
  • Security review + procurement support
  • On-premise / VPC deployment option
  • Compliance documentation (SOC 2 in progress)

Typical engagements start at $25K / year.

FAQ

What's an evaluation?
An evaluation is a single rule-check against a single agent output. Run a handful of rules against one trace, and you get one evaluation per rule. Traces themselves are free — we only meter the evaluations.
Why price on evaluations instead of traces?
Evaluations are the value unit. A trace with no rules applied is a log entry; a trace scored by 12 rules is an instrument. The Pro tier's 100K evaluations translates to roughly 8K traces with the default rule set, or more if you run a leaner subset.
Can I self-host?
Yes. The OSS MCP server (@iris-eval/mcp-server) runs locally today on npm + Docker; the dashboard + playground run in-process. The Cloud Starter tier (launching v0.4) adds hosted storage, scaling, and alerting on top of the same core. Enterprise VPC deployment combines the two.
What's the Self-Calibrating Eval beta?
A v0.5 feature that adjusts eval thresholds based on observed patterns in your traces, so rules stay useful as your agent evolves. Pro + Enterprise get early access. Details will ship with v0.5.
Is the Free tier forever?
The 10K evaluations / month threshold is the commitment we can make for individual use today. If your needs grow past it, Pro is priced to make the transition obvious.
How does billing work for Cloud Starter waitlist?
When Cloud Starter launches, waitlist members get first access + founding-member pricing lock for year 1. No payment until the tier goes live + you opt in.

Not sure which tier fits?

Start on Free. Move to Pro when your team's production agents exceed 10K evaluations a month (roughly 1K traces a day with default rules). Contact Enterprise when you need SSO, SLA, or custom deployment.

See the install docs →