Spanlens is an MIT-licensed observability platform for OpenAI, Anthropic, and Gemini workloads, offering request logging, cost tracking, and agent tracing. Teams integrate with a one-line SDK change, a drop-in proxy, or a zero-config CLI, and can self-host the same code with one Docker command.[1][2]
Quick verdict
Spanlens is a young MIT-licensed LLM observability platform that competes with Helicone and Langfuse on request logging, cost tracking, agent tracing, evals, and prompt management. Its distinctive offer is no feature gating by tier plus a free unlimited self-hosted option, at published prices below comparable hosted tiers. The main risks are project youth with 0.x SDK churn and a narrow set of native provider integrations.[1][2][3]
What Spanlens does
Spanlens captures every LLM call through a drop-in proxy for any language, one-line TypeScript and Python SDK changes, or a zero-config CLI init, with no vendor lock-in.[1][2]
Every request is logged with traces, sessions, users, cost, and latency, and agent runs are traced across LangChain, LangGraph, LlamaIndex, CrewAI, the Vercel AI SDK, and Bedrock.[1][2]
Every plan, including Free and self-hosted, includes request logging, cost tracking, agent tracing, prompt management, evals, anomaly detection, and PII and security scanning, with higher tiers adding capacity, seats, and integrations rather than core features.[1][2]
A versioned REST API with an OpenAPI specification exposes spans, traces, sessions, generations, scores, prompts, and datasets for programmatic access and automation.[1]
Who it fits
Teams running production LLM applications that need request-level logging and per-model cost visibility without re-instrumenting their code.[1][2]
Developers debugging multi-step AI agents built on LangChain, LangGraph, LlamaIndex, or CrewAI who need distributed traces of tool and retrieval steps.[1]
Cost-sensitive teams from solo developers on the Free tier to enterprises needing SSO/SAML, audit logs, and custom SLAs on Team and Enterprise plans.[1]
Limitations to weigh
Native provider support is limited to OpenAI, Anthropic, and Gemini; everything else must come through OpenTelemetry OTLP/HTTP ingest, which is less turnkey than a native integration.[1]
The project is young and evolving fast, with 9 of 10 planned releases shipped and breaking 0.x SDK changes documented in October 2026, so teams should expect API churn in the short term.[1]
Cloud data regions, offline use, and desktop or mobile platform support are not documented, so regulated or offline-first teams need to verify these gaps, most likely by self-hosting.[1]
Workloads centred on providers outside OpenAI, Anthropic, and Gemini depend on generic OTLP ingest rather than native integrations.[1]
Compatibility and competitive position
TypeScript SDK, Python SDK, and a language-agnostic proxy cover any stack, with OpenTelemetry compatibility for existing instrumentation.[1][2]
Self-hosting runs the same codebase as the hosted product via Docker, so teams can move between cloud and self-hosted without changing instrumentation.[1][2]
Spanlens positions itself directly against Helicone and Langfuse: its pricing sheet compares entry pricing with both hosted tiers at similar usage, and its team publishes a Helicone migration checklist for teams reconsidering their provider.[1][2]
Questions worth asking
Is Spanlens really free?
Yes, with two shapes. Self-hosting the MIT-licensed code with Docker costs nothing, has no request cap or seat limit, and runs the same code as the hosted product. The hosted cloud has a Free plan at 50,000 requests per month with 14-day retention and 1 seat, then Pro at $29 per month, Team at $149 per month, and custom Enterprise pricing.[1]
Which providers and frameworks does Spanlens support?
Spanlens natively supports OpenAI, Anthropic, and Gemini, and ingests anything else over OpenTelemetry OTLP/HTTP. Documented framework integrations include LangChain, LangGraph, LlamaIndex, CrewAI, the Vercel AI SDK, OpenAI Assistants, and Bedrock, plus a proxy mode that works with any language.[1]
How does Spanlens compare to Helicone and Langfuse?
Spanlens positions itself directly against both: its pricing sheet compares its entry pricing with Langfuse Cloud and Helicone hosted tiers at similar usage, and its team published a Helicone migration checklist. All three cover request logging, cost tracking, tracing, evals, and prompt management; Spanlens differentiates with no feature gating by tier and a free unlimited self-hosted option.[1][2]
Can I self-host Spanlens and keep my data?
Yes. Self-hosting is one Docker command with no request cap, and provider keys are AES-256-GCM encrypted server-side and never logged. Retention on self-hosted deployments is under operator control, while hosted retention ranges from 14 days on Free to 365 days or more on Team and Enterprise.[1][2][3]
Sources and editorial review
Source research. Reviewed on October 13, 2026. Written by Sanjib Sah. AI assisted; reviewed by a human editor.
We have not run or tested this product.
- spanlens/spanlens GitHub repository
official · Observed October 13, 2026
Repository identity, license, self-hosting, release history, and sponsored development.
- Spanlens AGENTS.md integration guide
official · Observed October 13, 2026
Machine-readable integration guide: positioning, integration paths, stack, frameworks, and license.
- Spanlens pricing fact sheet
official · Observed October 13, 2026
Machine-readable pricing: plan tiers, limits, SSO/SLA, self-host cost, competitor comparison.
- Spanlens Python SDK README
official · Observed October 13, 2026
SDK integrations, OTLP ingest, REST API surface, and SDK version history.
- Helicone acquisition migration checklist by Spanlens
community · Observed October 13, 2026
Official Spanlens account article positioning Spanlens as a Helicone migration target.
Compared with these products
No explicit proprietary baseline is recorded.