What is PostHog?
PostHog collects product events, session recordings, feature flags, errors, surveys and logs in one place, and adds AI observability on top: each model call is captured with its prompt, response, tokens, cost and latency, grouped into traces and linked to the user, their session replay and any exception. SDK wrappers cover OpenAI, Anthropic, the Claude Agent SDK, Google, the Vercel AI SDK, LangChain, LiteLLM and OpenRouter, among others.
PostHog AI is an analyst inside the app that writes SQL, builds insights and dashboards, finds replays and can open a coding task against a connected repository. It runs on Anthropic, OpenAI and Alphabet models and is available only in the cloud.
The repository is MIT-licensed outside the ee/ directory. Self-hosting is a free Docker Compose 'hobby' deployment that PostHog does not support; the company's own docs steer almost everyone to PostHog Cloud, which has a monthly free allowance on every product and usage-based pricing above it.
Is PostHog worth it?
PostHog's case for AI teams is context. A bad answer from your chatbot shows up as a trace with its prompt, cost and latency, and one click away sit the user's session replay, the funnel they dropped out of and the exception the request threw. Dedicated tracing tools such as Langfuse go deeper on prompt management and evaluation; none of them also run your feature flags and A/B tests. The repository has 40,826 GitHub stars and ships continuously from master rather than in tagged releases.
The cloud free tier is large: 1M analytics events, 5K recordings, 100K AI observability events shared with evaluations, and 500 PostHog AI credits worth $5 every month, and usage stops at the limits unless you add a card. Above the allowance, AI observability costs $0.00035 per event, falling to $0.00008 per event between 500,000 and 1M. By our arithmetic from those published rates, 1M AI events in a month costs about $100. Boost ($250 a month), Scale ($750) and Enterprise ($2,000) add longer retention, SSO and audit features.
Data handling needs reading. Privacy mode strips prompt and completion content from captured events, and PostHog removes large AI properties after 30 days. PostHog AI sends your messages and the data it reads to Anthropic, OpenAI and Alphabet; the DPA bars them from training on it, but there is no zero data retention, and PostHog trains its own models on anonymised data unless an admin switches that setting off.
Self-hosting is where the repository disappoints. The docs call it officially unsupported, offer no guarantees against data loss, publish no CVEs because there are no versions, and say all paid-plan features are cloud-only; PostHog AI cannot be self-hosted at all. The hobby deploy wants a machine of 4 vCPU and 16GB RAM, and the docs' own decision chart questions anyone expecting more than 300k events.
Where it earns its place
A five-person SaaS start-up adding an AI writing assistant to its app
See which users hit bad answers and what each conversation costs
Wrap the OpenAI or Anthropic client with PostHog's SDK so every call becomes a trace tied to the user and their replay. The first 100K AI events each month are free; 1M events costs about $100 at the published rates.
A product manager at a B2B software company
Ask questions of product data without waiting for an analyst
PostHog AI writes the SQL, builds the insight or dashboard and finds matching replays from a plain-English question. 500 credits a month are free, then $0.01 per credit; the data passes through Anthropic, OpenAI and Alphabet.
A healthcare app team with prompts that may contain patient details
Track model cost and latency without storing what patients typed
Turn on privacy mode in the SDK so prompt and completion fields are never captured, and sign PostHog's HIPAA BAA on the cloud. Token, cost and latency figures still arrive.
Before you commit
- Setup
- Cloud: sign up and run npx @posthog/wizard ai-observability in your project. Self-hosted: one deploy-hobby script on an Ubuntu VM with 4 vCPU, 16GB RAM and 30GB or more of storage, plus a domain.
- Integrations
- OpenAI, OpenAI Agents SDK, Anthropic, Claude Agent SDK, Google, Google ADK, Vercel AI SDK, LangChain, LiteLLM, OpenRouter, Cloudflare AI Gateway and more; an MCP server and Slack app; traces can also be forwarded from Helicone, Langfuse and Traceloop.
- Data and privacy
- Cloud regions in the US (Virginia) and EU (Frankfurt); HIPAA BAA available. Events are kept 1 year on the free plan and 2 years on pay-as-you-go. PostHog AI's providers may not train on your data, but PostHog trains its own models on anonymised data unless disabled.
- Cost at scale
- AI observability falls from $0.00035 to $0.00008 per event up to 1M; analytics events start at $0.00005 each after the first 1M. Platform packages run $250 to $2,000 a month on top. Self-hosted infrastructure costs are not published.
- Limits
- The free plan drops events beyond its limits. Self-hosted instances get no support, no tagged releases and no paid features, and new Kubernetes deployments are no longer supported.
What’s good
- LLM traces linked to session replays, errors and product analytics
- Free monthly allowance on every product, with usage capped by default
- SDK wrappers for most major model providers and agent frameworks
- Privacy mode keeps prompts and completions out of stored events
What’s not
- Self-hosted deployment is unsupported, unversioned and excludes paid features
- PostHog AI is cloud-only and has no zero data retention with its model providers
- Trains its own models on anonymised data unless an admin opts out
- Shallower prompt management and evaluation than dedicated LLM tools
Best for
Product teams shipping an AI feature who want model traces, cost and user behaviour in one cloud tool, and are content not to self-host.
PostHog alternatives
PostHog: common questions
- Is PostHog worth it?
- Worth it if you want LLM traces sitting next to the product analytics you already need, but treat it as a cloud service, because the self-hosted build is unsupported and leaves out the AI features. We rate it 4.3 out of 5.
- How much does PostHog cost?
- MIT outside ee/; self-hosted hobby deploy free and unsupported. Cloud free monthly: 1M events, 5K recordings, 100K AI observability events, 500 PostHog AI credits. AI observability then $0.00035 per event, falling with volume. Boost $250/month, Scale $750/month, Enterprise $2,000/month.
- What are the downsides of PostHog?
- Self-hosted deployment is unsupported, unversioned and excludes paid features; postHog AI is cloud-only and has no zero data retention with its model providers; trains its own models on anonymised data unless an admin opts out; shallower prompt management and evaluation than dedicated LLM tools.
- Who is PostHog best for?
- Product teams shipping an AI feature who want model traces, cost and user behaviour in one cloud tool, and are content not to self-host.
- How much does PostHog cost for a team or in production?
- AI observability falls from $0.00035 to $0.00008 per event up to 1M; analytics events start at $0.00005 each after the first 1M. Platform packages run $250 to $2,000 a month on top. Self-hosted infrastructure costs are not published.
- Is PostHog safe for company data?
- Cloud regions in the US (Virginia) and EU (Frankfurt); HIPAA BAA available. Events are kept 1 year on the free plan and 2 years on pay-as-you-go. PostHog AI's providers may not train on your data, but PostHog trains its own models on anonymised data unless disabled.
- What are the best PostHog alternatives?
- The closest alternatives we have reviewed are Langfuse, Arize Phoenix and Opik.
Answers checked Nov 16, 2026.