Model or dataset
Openpanel-dev/openpanel avatar
Openpanel-dev/openpanel

OpenPanel: what five containers, one encryption key, and a hosted MCP server actually commit you to

OpenPanel is an open-source web and product analytics platform, an open-source alternative to Mixpanel with optional self-hosting.

7,303 stars543 forksTypeScriptAGPL-3.0

At a glance

What is it?
OpenPanel is an open-source Mixpanel alternative with funnels, cohorts, session replay and self-hosting as a first class path. The self-hosted story is a specific stack rather than a vague promise: Postgres and Clickhouse split by workload, Redis forced to noeviction because it also holds queues, Redpanda in dev-container mode, and a single hex key covering TOTP, Search Console and object-store credentials.
Who is it for?
OpenPanel fits a team that wants funnels, cohorts, replay and revenue tracking without handing event data to a vendor, and that is willing to run five containers and own the Clickhouse side. Before committing, check four things.
Can I use it commercially?
Yes, with strict conditions. AGPL-3.0 is a network copyleft licence: if people use a modified version over a network, for example as a hosted service, you must offer them its source code under the same licence.
Is it still maintained?
Yes. The repository last received commits 6 days ago.
What is it written in?
Mainly TypeScript, according to GitHub's language statistics.

Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.

From daily snapshots

Star history

  • Openpanel-dev/openpanel7,303 ★ · +264
Star history
Repository2026-09-232026-11-15
Openpanel-dev/openpanel7k7.3k

264 Stars in the last 53 days, measured from the daily GitHub snapshot. Compare with other projects

Editorial analysis

Postgres holds the basics, Clickhouse holds the events

The stack is split by workload rather than by taste. Postgres stores basic information, Clickhouse stores events, and Fastify is the event API that receives them. The dashboard is Nextjs, and tRPC carries calls between the two sides. Redis sits alongside as the cache layer, pub/sub and queue at once.

That split shows up concretely in the environment file, where three connection strings sit side by side: DATABASE_URL with a schema=public parameter, DATABASE_URL_DIRECT defaulting to the same value, and CLICKHOUSE_URL pointing at localhost:23123 with the openpanel database.

Two stores also means two things to back up.

The consequence for anyone evaluating this is that a self-hosted install is not one database. It is Postgres, Clickhouse and Redis, and the high volume event data lives in the one that is hardest to inspect by hand.

The start sequence is containers first, then codegen, then one migration

Local development wants Docker, Docker Compose and Bun, and the root package.json requires bun >=1.4.4. The start sequence is fixed and the order matters, because the database does not exist until you create it:

bash
bun install
cp .env.example .env
echo "API_URL=http://localhost:3333" > apps/start/.env

bun run dock:up
bun run codegen
bun run migrate:deploy # once to setup the db
bun run dev

The migration step is marked as a one time database setup. After that the addresses are fixed: the dashboard answers on https://localhost:3000, the API on https://api.localhost:3333, and the queue's Bullboard on http://localhost:9999. Two terminals are provided for the stores, bun run dock:ch for a Clickhouse terminal and bun run dock:redis for a Redis terminal.

Redis runs with noeviction, so a full cache fails writes instead of dropping them

The Redis container starts with an explicit policy, redis-server --maxmemory-policy noeviction. With that policy Redis refuses writes when memory fills instead of discarding keys, which matters for a queue holding jobs.

OpenPanel uses Redis for three roles at once, listed in the stack as cache layer, pub/sub and queue, with BullMQ as the queue and GroupMQ as a grouped queue alongside it. That is why the policy matters. Had this instance held only cached dashboard results, eviction would be a performance decision. Since it also holds work, a full instance becomes a stuck pipeline rather than a slow page.

The ingest numbers sit near it in the environment file. BATCH_SIZE is 5000 and BATCH_INTERVAL is 10000 on the REST path with CONCURRENCY at 10, and WORKER_PORT is 9999, the port Bullboard exposes. If you change your memory expectations on a self-hosted install, this policy is the first line to revisit.

Redpanda runs in dev-container mode and its auth profile is behind a flag

Event streaming runs on Redpanda, configured as redpandadata/redpanda:v25.1.4. Two listeners are advertised: an internal one on port 9092 addressed as op-rp, and an external one on 19092 advertised as localhost:23092, so containers reach the broker by service name while your machine reaches it on the mapped port. The RPC listener sits on 33145.

The resource settings are deliberately small, with --smp=1, --memory=1G and --mode=dev-container. That mode is a development convenience and also a boundary: the broker is configured not to behave like a production cluster.

One more tuning hint: the Clickhouse container maps its HTTP interface (8123), native TCP (9000) and inter-server port (9009) to host ports 23123, 23900 and 23909 on 127.0.0.1, and raises its file descriptor limit to 262144 in both soft and hard form.

One hex key encrypts TOTP secrets, Search Console tokens and export credentials

At-rest encryption in a self-hosted install runs on a single value, ENCRYPTION_KEY, generated with openssl rand -hex 32. The comment above it is specific about the blast radius: TOTP secrets, Search Console tokens, and object-store export credentials, meaning S3 secret keys and GCS service-account keys.

The same file documents the optional export path. A flushExports cron job windows Clickhouse by inserted_at and uploads batched files, with a lag setting holding a safety gap behind now() so in-flight inserts are not missed, a batch size setting rows per file, a max batches per run value bounding backlog drain per tick, and a concurrency value setting parallel uploads per project and integration. The example values are 60, 50000, 20 and 4. Each knob changes either how much data moves or how hard your object store is hit, and all four are optional.

Session replay has no session cap here, and the table is where that is said

The feature list puts session replay alongside funnels, cohorts and user profiles, with privacy controls described as built in, and the deployment story is cookieless tracking with GDPR compliance.

The comparison table is where the concrete difference sits. Under session replay, the note says Mixpanel is limited to 5k sessions a month on its free tier and 20k on paid, while OpenPanel has no limit. The same table marks user profiles and session history as present in OpenPanel and Mixpanel and absent in GA4 and Plausible, and event and funnel notifications as present in the first two and absent in the last two. Plausible is annotated as having simple goals rather than the full funnel model.

Read those cells as positioning rather than measurement. They are the project's own table, so treat a competitor's missing feature as a claim to check against that product's current documentation before you move a team.

What is verifiable here is the shape of the claim. Unlimited replay and self-hosting are what the project is selling, and both only pay off if you actually intend to run the stack.

MCP is 38 tools on a hosted instance, so it points away from your own data

One entry in the feature list does not behave like the others. The MCP server is described as hosted, with no install, and the line asks whether Claude, Cursor or any MCP client can answer questions about your users against it. Thirty-eight tools is the count given.

That positioning has consequences for a self-hosted deployment. Self-hosting is offered as an option, and a self-hosted install has to be stood up by you. The MCP server is presented as something that already exists. If your events live in your own Postgres and Clickhouse, tools aimed at a hosted instance are aimed at data you do not control, so the one capability that needs no install is the one that is hardest to point at an internal deployment.

The README says nothing further about it.

For evaluating the project, split the list into two questions: which capabilities will you run yourself, and which one will you keep calling across the network.

The pre-push hook runs typecheck and vitest, and one variable turns it off

The root package.json's test script chains bun run test:core and bun run test:rest, and the tree carries vitest.config.ts, vitest.workspace.ts and a root biome.json.

Two things in the tree deserve a look before you build on it. License metadata at the repository level is AGPL-3.0, while the private root package.json declares MIT, so if you intend to vendor or redistribute pieces you need to know which terms apply. The repository also ships SECURITY.md and TRADEMARK.md next to LICENSE.md, alongside a self-hosting directory, which is a signal that running your own instance under your own brand is a scenario the project has thought about.

Editorial conclusion

OpenPanel fits a team that wants funnels, cohorts, replay and revenue tracking without handing event data to a vendor, and that is willing to run five containers and own the Clickhouse side. Before committing, check four things. Whether the unlimited session replay claim matters to you, since the comparison table is the project's own. Which license governs what you vendor, because the repository is AGPL-3.0 while the private root package.json declares MIT. Whether ENCRYPTION_KEY and its blast radius fit your threat model, since one key covers TOTP secrets, Search Console tokens and export credentials. And whether Redpanda in dev-container mode is a starting point you are willing to replace.

Frequently asked questions

what is openpanel

OpenPanel is an open-source web and product analytics platform positioned as an alternative to Mixpanel, covering funnels, cohorts, user profiles, session replay, A/B testing and revenue tracking.

how to install openpanel

For self-hosting, follow the guide at the OpenPanel docs, or for local development install Docker, Docker Compose and Bun, then run bun install, copy .env.example to .env, start the containers with bun run dock:up, run bun run codegen and bun run migrate:deploy, and finish with bun run dev.

is openpanel free

The project describes transparent pricing with no hidden costs or usage limits, and self-hosting is supported, and the MCP server is offered as a hosted service.

Official sources

  1. Issues
  2. License: AGPL-3.0
  3. Openpanel-dev/openpanel on GitHub
  4. Project website
  5. README
Add this badge to your README

If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.

Add this badge to your README

markdown
[![Hysen Labs](https://hysenlabs.com/badge/openpanel-dev-openpanel.svg)](https://hysenlabs.com/projects/openpanel-dev-openpanel)

Star history chart for your README

markdown
[![Star history](https://hysenlabs.com/chart/openpanel-dev-openpanel.svg)](https://hysenlabs.com/projects/openpanel-dev-openpanel)

Was this useful?