Logo

Spanlens

Open-source LLM observability: log cost, latency, tokens, an

Spanlens

Spanlens

Open-source LLM observability: log cost, latency, tokens, an

Screenshots

Screenshot 1
Freemium
AI, Developer Platform
Built with
.NET.NET

Spanlens is an open-source (MIT) LLM observability platform that lets developers monitor every call their application makes to OpenAI, Anthropic, Gemini, Mistral, OpenRouter, Azure OpenAI, or a local Ollama model. Integration takes one line: swap your client's baseURL to the Spanlens proxy, or run "npx @spanlens/cli init" and the wizard rewrites your code automatically. From that moment, every request is recorded with its model, token counts, latency, cost, and full prompt and response body, with streaming responses reconstructed automatically.

The dashboard turns that raw log into operational insight. Cost tracking breaks spend down per request, per model, and per end user, and parses prompt-cache tokens separately so you see real cache savings rather than sticker price. Agent tracing visualizes multi-step workflows as Gantt waterfalls and node-and-edge graphs, highlighting the critical path so you can find the slowest dependency chain in a fan-out. Anomaly detection flags 3-sigma deviations in latency, cost, or error rate against a rolling 7-day baseline with root-cause hints. Alerts on budget, error rate, and p95 latency are delivered to Email, Slack, or Discord.

Spanlens goes beyond passive logging. A regex-based PII and prompt-injection scanner inspects request and response bodies and can block injections at the proxy. The savings engine spots calls that match a cheaper model's profile (for example, a gpt-4o call that looks like a classification task) and estimates the monthly saving from switching. Prompt versioning with A/B experiments compares versions on latency, cost, and error rate using Welch's t-test for statistical significance, and an LLM-as-judge evaluation framework (judge with OpenAI, Anthropic, or Gemini) scores outputs against rubric anchors, with human agreement measured by Pearson r or Cohen's kappa. Reusable datasets power offline evals and regression checks.

Related products

View all
AntForms

AntForms

Unlimited free submissions + free analytics + integrations +

7
Forms, AI, Productivity
Huntopic

Huntopic

AI that finds customers for you on Reddit

14
Marketing, Sales, AI
AthleteMatrix

AthleteMatrix

Your personal coach based on data from your everyday apps

2
AI, Productivity, Fitness
OneTwoResume

OneTwoResume

OneTwo Resume is an AI-powered resume builder that helps job

0
AI
Tenth Man

Tenth Man

AI built to disagree

1
AI, Data Analytics
git-lrc

git-lrc

Free, Unlimited AI Code Reviews That Run on Every Commit

3
AI, Code Review, DevTool

Launched in Week 38, 2026

See the week
DevCleaner

DevCleaner

Reclaim Gigabytes from Your Mac Dev Tools

2
DevTool, App Development, Automation
InviteVia

InviteVia

Create beautiful digital invitations and collect RSVPs.

0
Marketing
SimpleTradeLog

SimpleTradeLog

Import your Interactive Brokers trades, see your real net P&

0
Finance
Thefake

Thefake

Create realistic chats, posts & videos in your browser

0
SaaS
PurrPlan

PurrPlan

AI-powered social media management from €9/month

0
Social Media Management
ConsentStack

ConsentStack

Consent Management Platform, CMP, Website cookies, Cookie ba

0
AI