PostHog
Open-source product platform with LLM analytics (traces, cost, latency, evals), session replay, feature flags and a built-in AI analyst.
40.5k
GitHub Stars
100K
Free LLM Events / Month
2020
Founded (YC W20)
Overview
PostHog is an open-source product platform that bundles product analytics, session replay, feature flags, experiments, error tracking, surveys and a data warehouse. For AI teams, its LLM analytics product records every model call (prompt, response, tokens, cost, latency, tools used), groups calls into traces, and links each trace to the user, their session replay and any exceptions. PostHog AI, the built-in analyst, answers questions in plain English, writes SQL and builds dashboards, and an MCP server brings the same data into Claude Code, Cursor and other agents.
The Verdict
Who Should Use PostHog?
Best For
- Teams shipping LLM features who want AI traces next to product analytics
- Startups that want one tool instead of separate analytics, replay and flags vendors
- Engineers who want to query product data from Claude Code or Cursor via MCP
- Cost-conscious teams that fit inside the generous monthly free tiers
Not Ideal For
- Teams that only need deep LLM evals and prompt management (a dedicated tool goes further)
- Non-technical users who need reports without setup work
- Large companies wanting a supported self-hosted deployment (open-source self-hosting is hobby-scale, without support)
What's Great
- LLM traces link to the user, session replay and errors, which pure LLM tools can't do
- Integrations with about 50 SDKs and frameworks, from OpenAI and Anthropic to LangChain, Vercel AI SDK and Claude Code
- Imports traces from Helicone, Langfuse and Traceloop
- Free tier on every product, kept even on paid plans
- Reviewers praise easy setup and useful session replays
Watch Out For
- Learning curve is the most common review complaint
- So many products that new users report feature overload
- Event and dashboard setup needs planning; some reports are hard for non-technical users
- Usage-based pricing across many products makes bills harder to predict
- Self-hosting is a Docker "hobby" deploy, recommended only up to about 100k events a month
Pricing
Free tiers
$0
Every month: 1M analytics events, 5K replays, 1M flag requests, 100K LLM events, 500 PostHog AI credits
Pay as you go
Usage-based
LLM analytics from $0.00035/event, PostHog AI $0.01/credit, volume discounts
Platform add-ons
$250-$2,000/mo
Boost $250, Scale $750, Enterprise $2,000, on top of usage
View all features & details
LLM Analytics
- Captures prompt, response, tokens, cost, latency and tool calls
- Traces linked to users, replays and exceptions
- Cost and latency dashboards
- Failure clustering and a prompt playground
- Evaluations and user feedback on traces
- Access via web app, MCP server or API
PostHog AI
- Plain-English questions over your product data
- Writes SQL and builds insights and dashboards
- Creates surveys, alerts and feature flags
- Can start coding tasks on a connected repo that end in a PR
- Project memory across conversations; Slack access in beta
Other Products
- Product analytics and web analytics
- Session replay (web and mobile)
- Feature flags and experiments
- Error tracking, logs and surveys
- Data warehouse
Open Source
- MIT license, except the
eefolder - posthog-foss variant with proprietary code removed
- Hosted MCP server for Claude Code, Cursor, Codex, VS Code and Zed
How It Compares
| Feature | PostHog | Langfuse | LangSmith |
|---|---|---|---|
| Focus | Product platform + LLM analytics | LLM engineering | LLM apps, LangChain |
| Open source | Yes (MIT core) | Yes (MIT) | No |
| Free LLM tier | 100K events/mo | 50K obs/mo | 5K traces/mo |
| Links traces to replays and users | Yes | No | No |
| Prompt management | Playground | Built-in | Built-in |
| Self-hosting | Hobby deploy only | Free, supported | Enterprise |
User Reviews
Loading reviews...