What it costs to run agents you can actually defend.
Same platform at every tier — pay for scale, team tooling, and runtime policy controls.
Prices are in Australian dollars and exclude GST, added at checkout for Australian customers. Your card is charged in AUD, converted by your bank at their rate on the day. Plans renew automatically until cancelled — cancel any time from Settings, and access runs to the end of the period you have paid for.
Self-host forever, or stay free on our cloud. Single workspace.
- Full SDK + self-hostable core
- Observe · replay · signed audit
- CI release gate · PII redaction
- Any model · OpenTelemetry compatible
- Single workspace · 1,000 runs/mo · 7-day retention
Small team, managed — no infra to run.
- Everything in Community, hosted
- Team & roles (RBAC) · 5 seats
- Alerting — email · Slack · webhook
- 20,000 runs/mo · 30-day retention
Teams with agents in production.
- Everything in Starter · 10 seats · 50k runs
- Golden corpus — incidents auto-mined into regression tests
- Eval runner + CI gate · dataset management
- Versioned prompt registry + playground
- Pairwise eval comparison + judge calibration
- 60-day retention · multi-workspace
Multi-team, fleet visibility, policy control.
- Everything in Growth · 20 seats · 75k runs
- Fleet control-room dashboard
- Policy simulation + library
- Model diff · upgrade gate · cost attribution
- Policy causal attribution heat-map
Full platform, named support, managed SLA.
- Everything in Scale · 25 seats · 100k runs · 90-day retention
- Fleet benchmarks vs. vertical peers
- Inter-agent trust chain · signed delegation proofs
- Live runtime enforcement · advanced bisection
- Priority support · named contact · response targets
For regulated environments.
- Everything in Pro, in your perimeter
- SSO (OIDC) · managed audit keys · legal holds
- Unmetered runs · unlimited retention
- Self-host in your VPC · data residency
- Regulatory dashboard — EU AI Act · ISO 42001 · NIST AI RMF · APRA CPS 230/234 · GDPR · ISO 27001
- Team chargeback with monthly budget caps
- Named support · SLAs · security review
Self-hosting is available on Community or Enterprise. Named regulatory mappings (EU AI Act, ISO 42001, NIST AI RMF, APRA CPS 230/234, GDPR, ISO 27001) are Enterprise only — status computed live from your run data, not a static crosswalk — see the comparison below.
What you get at each tier.
Swipe to compare all six tiers →
| Capability | Community | Starter | Growth | Scale | Pro | Enterprise |
|---|---|---|---|---|---|---|
| Capture · replay · signed audit | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
| In-process PII redaction · any model · OTel | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
| Evals — run against a dataset | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
| Seats | 1 | 5 | 10 | 20 | 25 | Unlimited |
| Runs / month | 1,000 | 20,000 | 50,000 | 75,000 | 100,000 | Unlimited |
| Retention | 7 days | 30 days | 60 days | 60 days | 90 days | Unlimited |
| Team roles (RBAC), alerting & approvals | — | ✓ | ✓ | ✓ | ✓ | ✓ |
| Golden corpus, prompt registry, pairwise eval judging & incidents | — | — | ✓ | ✓ | ✓ | ✓ |
| Fleet dashboard, policy simulation, model diff/drift detection & cost attribution | — | — | — | ✓ | ✓ | ✓ |
| Fleet benchmarks, anomaly signals & named support/SLA | — | — | — | — | ✓ | ✓ |
| Runtime enforcement, SSO, chargeback & regulatory mappings | — | — | — | — | — | ✓ |
| Self-host in your perimeter | ✓ | — | — | — | — | ✓ |
| Data residency | Your infra or US cloud | US cloud | US cloud | US cloud | US cloud | Your perimeter |
| Support | Community | Priority email | Priority email | Named contact · response targets | Named contact · contracted SLA |
Named regulatory frameworks are detailed in Regulatory dashboard & compliance export; API rate limits per tier are in Rate limits.
Community edition — free to self-host.
The source is public — clone it and it's one command to deploy. Or skip the infra and start managed at A$49/mo with a free trial.
The questions that come up before signing.
We already use LangSmith or Langfuse — why switch?
You don't have to. Those tools make traces readable; Runback makes them re-executable — re-run any step from the exact captured context, gate a model upgrade in CI, export a signed artifact your auditor can verify independently. Most teams run both. See the full breakdown: Runback vs. LangSmith · Runback vs. Langfuse.
Where does our data actually go?
Community: your Postgres, your infra — nothing leaves your perimeter. Starter through Pro: Runback-managed Postgres (US region), PII redacted in-process before anything is sent. Enterprise: your Postgres in your own VPC. Residency outside the US (EU, AU) requires Enterprise self-hosted.
What happens if Runback is down during policy enforcement?
Observation degrades gracefully — a failed trace send retries in-process on the next flush, though that buffer isn't persisted, so events not yet flushed are lost if the process exits first. Policy gates fail open by default, so your agent is never blocked by an outage. Enterprise can configure fail-closed where blocking is the safer default.
Can't we just build this ourselves?
The capture layer is a sprint. The parts that compound — policy simulation over 90 days of real decisions, a golden corpus that auto-mines regressions, fleet benchmarks that only mean something at scale — take a year and don't work without the data. Your logging history can't be back-filled.
Is the Community edition really free?
Yes, free to self-host forever, and the source is public — clone runback-community, no request needed. Starter (A$49/mo) gets your team on managed hosting with RBAC and alerting. Growth (A$600/mo) is where compound value starts — the golden corpus builds your regression library from real failures, automatically.
What's the difference between free and paid?
Community gives you capture, replay, signed audit, and the CI gate — free under the Runback Community License. Enterprise features (team roles, SSO, fleet dashboard, policy enforcement) need a plan or signed self-host license. An arbitrary RUNBACK_LICENSE string unlocks nothing — it's cryptographically gated.
How do we get through a security review?
Start with one agent in your own VPC — nothing leaves your perimeter, so most reviews are short. We'll join the call and answer questions directly. The procurement kit has the documents your assessment needs.
Can your team read our agent traces?
On managed cloud: infrastructure staff have database-level access. PII is redacted inside your application process before anything arrives — card numbers, emails, and SSNs never reach us — but redacted trace content is readable by anyone with database access. If that's not acceptable, self-host: your data never touches our servers.
What happens to our audit records if Runback shuts down?
Self-hosted customers own the database — records stay in your infrastructure regardless of what happens to Runback. Managed cloud: cassette files use an open, documented format (runback.cassette/v1) with a public schema, exportable any time and verifiable independently, without Runback software or an account.
GDPR right to erasure vs. your append-only ledger — how does that work?
A real tension. The append-only ledger is designed so removing an entry breaks the hash chain for all subsequent entries — that's the point of tamper-evidence. Our approach: configure redaction so personal data never reaches the cassette. If PII did reach a managed-cloud cassette and a verified erasure request arrives, we delete the run record — the break is the honest record of what happened. Self-hosted: you own the database and the legal decision.