A Langfuse alternative, compared honestly
Acquired by ClickHouse on 16 January 2026; open source, cloud and team continue. Langfuse is a tracing and evaluation platform and Vigil is a cost proxy: they overlap on cost per call and little else, so this page is for the reader who used Langfuse mainly to watch spend. If you are looking at alternatives to Langfuse, this page puts Vigil beside it without flattering either. Every Langfuse fact was read on Langfuse's own pages on 2026-10-09, its pricing and status again on 2026-11-16, and links to its source. Vigil does less than Langfuse in several places, listed plainly below; what it does differently is a proxy that creates prompt-cache hits and prices every call with its provenance shown.
What Langfuse is, and what Vigil is
Langfuse, in its own words: “Trace, evaluate, and improve AI agents with one open platform.”langfuse.com. SDK, outside the request path: "from langfuse.openai import openai"; traces are sent asynchronously. Re-read 2026-11-16: an AI Gateway "in public beta" now also runs in the request path, "No retries, fallbacks, or caching".
Vigil is a proxy for AI API calls. Change one base URL, add one header, and every call is logged with its cost, tokens, latency, errors and agent, then forwarded unchanged unless you switch on an optimisation. Prompt caching, where the provider supports it, never changes answers. History trimming and model routing can change answers: both are opt-in, run in Shadow first, and stop automatically if errors, cut-off answers or failed answers rise. It stores no prompt text and caches no responses.
- Langfuse says: "No, Langfuse is not an LLM Proxy." It is "an asynchronous observability layer", to avoid "introducing a single point of failure".langfuse.com
What happened to Langfuse
ClickHouse announced on 16 January 2026, with its Series D: "ClickHouse is thrilled to announce the acquisition of Langfuse, the open-source LLM observability platform." Its companion post says "Langfuse remains 100% open-source under its existing MIT license for core features" and "Langfuse Cloud continues operating as a standalone service." Langfuse’s own announcement the same day: "Langfuse stays open source and self‑hostable." and "Langfuse Cloud keeps running as‑is."; its press page lists "Acquired by ClickHouse: January 2026". On 16 November 2026 the pricing tiers were unchanged and the README read "Since January 2026 we’re part of ClickHouse."
Sources, read 2026-11-16: ClickHouse, 16 January 2026 · ClickHouse, "ClickHouse welcomes Langfuse" · Langfuse, 16 January 2026
Feature by feature
The Langfuse column was read on Langfuse's own pages on 2026-10-09, and the cells marked “re-read” again on 2026-11-16; each cell links to the page. “Not stated” means the pages read do not mention it, which is not the same as “no”. The Vigil column is read from Vigil's code and docs.
| Feature | Langfuse | Vigil |
|---|---|---|
| Integration | SDK and OpenTelemetry; "Fully async requests, meaning Langfuse adds almost no latency."langfuse.com | Proxy. Change the base URL and add the X-Vigil-Key header; no SDK. |
| Open source | "MIT licensed, except for the ee folders."github.com | No. Vigil is not open source. |
| Hosting | Cloud in US, EU, Japan and a HIPAA region; self-host by Docker Compose, Helm or Terraform.langfuse.com | Cloud only: a Cloudflare Worker in front of the provider, a dashboard on Vercel, data in Supabase (Frankfurt, EU). |
| Cost per call | Yes, inferred from the model name and ingested usage: "Langfuse works them out from the generation’s model parameter"; custom models need a definition.langfuse.com | Yes: cost, tokens, latency, status and agent on every call, from a rate registry where every rate carries its source, date and provenance; a dash where no rate is published. |
| Providers | "100+ integrations". Cost needs a model definition, not a proxy route.langfuse.com | 17 providers routed by segment. |
| Prompt caching | Reports cache-read tokens from the usage your SDK sends. It is not in the request path, so it cannot create a hit.langfuse.com | Creates cache hits: Anthropic, Google Vertex, Mistral get cache markers placed by Vigil when optimisation is on; AWS Bedrock when the request carries an API key rather than a SigV4 signature. On other providers Vigil records the provider’s own caching. |
| Response caching | None for LLM responses; prompts fetched from prompt management are cached client-side. The beta AI Gateway, re-read 2026-11-16: "No retries, fallbacks, or caching."langfuse.com | No. Vigil does not cache responses. |
| Routing and fallbacks | None today: "Langfuse does not sit between your application and the LLM provider’s API"; the beta AI Gateway, re-read 2026-11-16, lists "Model routing" and "fallbacks between providers" as roadmap.langfuse.com | Model routing is in beta: Sends a call to a cheaper model from the same provider — one tier down — when its answer is closed-form and checkable (a label, JSON or other data, a number, code to a named function or test, or a request with tools), its prompt is at most about 16,000 tokens, its opening's size and tools are in a band the agent's routing level covers and the cheaper model can take the request. Open-ended work without tools — writing, summaries, reviews, explanations, chat — stays on the original model, and once a call of a conversation stays there, the later ones do too; a later call the cheaper model can't take stays on the original model. Routing can change answers, so it runs in Shadow first. No provider fallbacks or load balancing. |
| Prompt management | Yes: "storing, versioning, and retrieving prompts", labels, rollbacks.langfuse.com | No prompt management or versioning. |
| Evals | Yes: LLM-as-a-judge, scores, annotation queues, datasets and experiments.langfuse.com | No evaluations. A score can be sent for a call through the feedback endpoint. |
| Tracing | "hierarchical traces capture every LLM call, tool invocation, and retrieval step".langfuse.com | One record per call. No spans or traces. |
| Alerts | Metric alerts to Slack, webhooks and GitHub Actions; 2, 20, 50 or 100 per plan.langfuse.com | Cost spikes, failures and findings are flagged in the dashboard. Nothing is sent by email, Slack or push. |
| Prompt injection and PII | No detection. Masking functions you write "redact sensitive information before trace data leaves your application".langfuse.com | Flags likely prompt injections and personal data in prompts, on every plan. Nothing is blocked or redacted. |
| Budgets and rate limits | None on LLM spend (not in the request path); ingestion rate limits per plan.langfuse.com | No spend caps or rate limits on your provider traffic. Optimisation can be switched off per agent. |
| Seats | 2 users on Hobby, unlimited from Core; project-level RBAC with the Teams add-on or Enterprise.langfuse.com | Seats per plan: 1 on Free, 3 on Plus, unlimited from Pro. Teammates are invited; there are no role tiers. |
| Compliance | "covered by SOC 2 Type II and ISO 27001 audits"; GDPR; a HIPAA-ready region.langfuse.com | No SOC 2 or ISO 27001 report. A DPA is published. |
| Data residency | US (us-west-2), EU (eu-west-1, Ireland), Japan (ap-northeast-1).langfuse.com | Account and telemetry stored in the EU (Supabase, Frankfurt). The proxy runs on Cloudflare’s edge. No prompt text or responses are stored. |
Where Langfuse is stronger
- Tracing with spans across every LLM call, tool invocation and retrieval step.
- Evaluation: LLM-as-a-judge, annotation queues, datasets and experiments.
- Prompt management with versioning, labels and rollbacks.
- Open source under MIT, self-hostable, with SOC 2 Type II, ISO 27001 and a choice of three regions.
- Outside the request path, so it cannot add latency or fail a request.
Where Vigil differs
- It creates prompt-cache hits rather than reporting them, on the providers whose caching lets a proxy place markers, and measures what caching would have saved on the others.
- Every rate it prices with carries its source, its retrieval date and a provenance label, and a cost it cannot stand behind is a dash, not a number.
- Monitoring is not gated by plan: every agent, every provider, on Free. Paid plans switch optimisation on.
- It does less: no response cache, no evals, no prompt management, no traces, no alerts outside the dashboard.
Pricing, side by side
Langfuselangfuse.com read 2026-11-16
- Hobby Free: 50k units a month (a unit is a trace, an observation or a score), 30 days of data access, 2 users
- Core $29/mo: 100k units, $8 per further 100k; 90 days of data access; unlimited users
- Pro $199/mo: 100k units, 3 years of data access, SOC 2 and ISO 27001 reports; Teams add-on $300/mo for SSO and RBAC
- Enterprise $2,499/mo: 100k units, audit logs, SCIM, SLAs; custom volume pricing
- Self-host Free: "all core Langfuse features for free without any limitations"; enterprise edition by contract
Tiers re-read 2026-11-16: unchanged; the overage steps down from $8 to $6 per 100k units at volume.
Vigil vigil.tools/pricing
- Free $0: every agent monitored; 10,000 logged calls a month, 7 days of history; every optimisation On until $30 saved, once ever, then a 30-day grace, then Shadow only
- Plus $29/mo: every optimisation On up to $100 saved a billing month, then a 30-day grace, then Shadow; 500,000 logged calls, 30 days
- Pro $99/mo: every optimisation On up to $500 saved a billing month, then a 30-day grace, then Shadow; 2,000,000 logged calls, 90 days
- Supreme $299/mo: every optimisation On up to $2,000 saved a billing month, then a 30-day grace, then Shadow; Unlimited logged calls, 365 days
- Enterprise Custom: by contract
Monthly, excl. VAT. Monitoring is free on every plan; what a paid plan buys is a higher limit on the money Vigil saves you per billing period before optimisations go to Shadow (after a 30-day grace).
Moving from Langfuse: what changes and what you keep
Vigil is a base-URL change and one header, so there is nothing to uninstall and nothing to instrument: the SDK you already use keeps its code. Your provider keys stay where they are. If you depend on Langfuse for a response cache, routing, prompt management or traces, Vigil does not replace those; the table above says which. Vigil is not open source and runs only as a hosted service, so a self-hosting requirement rules it out.
What you gain is a proxy that can place prompt-cache markers for you on the providers that allow it, a cost on every call with its provenance shown, and monitoring that is free on every plan. Start on Free and compare your own numbers before switching anything off.
Questions
- Is Langfuse still maintained?
- Yes. ClickHouse acquired Langfuse on 16 January 2026 and both companies said the product continues: "Langfuse remains 100% open-source under its existing MIT license for core features", "Langfuse Cloud continues operating as a standalone service", and from Langfuse, "Our roadmap stays the same". On 16 November 2026 the pricing tiers were unchanged.
- Who acquired Langfuse?
- ClickHouse, the database company, announced the acquisition on 16 January 2026 alongside a $400 million Series D; the price paid for Langfuse was not disclosed on either company’s pages.
- Does Vigil replace Langfuse?
- No, for most of what Langfuse does: traces, evaluations, prompt management and datasets are not things Vigil has. Vigil is a proxy that prices every call in the request path and can create prompt-cache hits, which Langfuse, outside the request path by default, cannot; teams use the two for different jobs.
Also: Helicone alternative · Portkey alternative · LiteLLM alternative · OpenRouter alternative