Observability
Bonito vs Langfuse
Open-source LLM observability vs. a control plane with observability built in
Acquired by ClickHouse (January 2026); remains open source
Langfuse is an open-source platform for LLM tracing, evals and prompt management, now owned by ClickHouse. It observes calls made elsewhere and is deliberately not a proxy. Bonito sits in the request path: it routes, fails over, governs and runs agents, and its trace tree comes with the calls it makes.
Langfuse pricing
Hobby free (50K units/mo); Core $29/mo; Pro $199/mo; Teams add-on $300/mo; Enterprise $2,499/mo. Self-hosting free.
Model costs
Not applicable: Langfuse is not in the request path and does not bill inference.
What Langfuse does well
- Deep open-source tracing, evals and prompt management, free to self-host
- Cost and token tracking by model, user and tag across providers
- Compliance: SOC 2 Type II, ISO 27001, GDPR and HIPAA BAA, with US, EU and Japan regions
- Wide adoption and an official hosted MCP server
- Works with any provider or gateway through SDKs and OpenTelemetry
Where Bonito goes further
- An OpenAI-compatible gateway across eight providers on your own accounts with 0% markup
- Automatic failover to an equivalent model on another provider, and cost routing to cheaper models
- Builds and runs agents: visual canvas, chat builder, approval queue, schedules and persistent memory
- Enforcement, not only observation: per-agent dollar budgets, default-deny tool policies and model allow-lists
- Managed knowledge bases with a BYO bucket
- Data residency pinning to CA, US or EU for the model calls themselves
Where Langfuse is ahead of Bonito
- Open source and free to self-host
- Eval depth and prompt management
- Framework-neutral tracing of calls that do not go through Bonito
Capability by capability
| Capability | Bonito | Langfuse |
|---|---|---|
| Agents | ||
| Build and run agents | Yes | Notraces agents; does not build or host them |
| Approval queue and scheduled runs | Yes | No |
| Dollar budgets per agent | Yes | No |
| MCP support | Yes | Yes |
| Routing | ||
| OpenAI-compatible gateway | Yes | Nonot a proxy, by design |
| Automatic failover across providers | Yes | No |
| Cheaper models for simple work (cost routing) | Yes | No |
| Bedrock, Vertex, Azure and direct providers in one place | Yes | Partialobserves cost across providers; does not route |
| Semantic caching | No | Nonot a proxy |
| Governance | ||
| SSO, RBAC and audit log | Yes | YesSSO and RBAC via Teams add-on; audit logs on Enterprise |
| Data residency pinning | Yes | PartialUS, EU or Japan regions for Langfuse data |
| Observability and knowledge | ||
| Agent traces with cost per request and agent | Yes | Yes |
| Knowledge bases (RAG) | Yes | Nonot found in their docs (Sep 2026) |
| How you run it | ||
| Whole stack from one config file | Yes | Nonot found in their docs (Sep 2026) |
| Self-hosted or open source | Partial | Yes |
| No markup on inference | Yes | Yesnot in the request path |
Checked against Langfuse's own documentation on 2026-09-27. Products change quickly; if something here is out of date, tell us at hello@trybonito.com and we will fix it.
Sources
See Bonito on your own stack
Start free, or get a report on where agents fit in your company first.