# Rawbbit (rawbbit.one)

> Rawbbit is open-source game analytics infrastructure that collects player events, preserves the raw history in a portable format, and makes the data available through dashboards, SQL, and MCP-connected AI agents.

Last updated: 2026-08-08

## Product definition

Rawbbit is designed for game studios that need useful analytics but do not want their event history trapped inside a hosted dashboard vendor. It provides the ingestion, raw storage, analytical database, dashboards, and AI-agent access layer as one deployable stack.

The central design rule is that raw Parquet remains the system of record. ClickHouse, data models, dashboards, and agent-facing tools sit downstream and can be rebuilt without changing event ingestion.

## Event ingestion contract

Games and backends send event batches to the Collector API over HTTP using a per-project API key. Each event has a consistent envelope with these required fields:

- `event_id`
- `app_id`
- `event_name`
- `event_timestamp`

Event-specific details remain flexible inside the event parameters. The collector validates the envelope and basic structure, enriches accepted events with technical metadata, and either accepts the batch or rejects it with a reason.

The HTTP endpoint is available today for game clients and backends. Native Unity, Unreal, and Godot SDKs are planned.

## Current architecture

The supported production path is:

1. A game client or backend sends event batches over HTTP.
2. The Collector API authenticates, validates, and enriches accepted events.
3. NATS JetStream buffers the write path and provides at-least-once delivery.
4. The raw writer lands partitioned Parquet files in S3-compatible object storage. Open-source SeaweedFS is the default deployment, and GCS is also supported.
5. A scheduled dbt Core runner loads bounded Parquet windows into the `analytics.events` table in ClickHouse and runs data tests.
6. Metabase dashboards, direct SQL users, and the Rawbbit MCP server read from ClickHouse.
7. Codex, OpenCode, OpenClaw, Claude, or another MCP client can use the read-only analytical interface to answer questions in plain language.

## Data ownership and portability

- Raw events land as open-format Parquet in object storage controlled by the studio, unless Rawbbit hosting is explicitly requested.
- The Parquet layer is the durable system-of-record boundary.
- ClickHouse is the fast serving layer and can be rebuilt from the raw files.
- The stack is open source under the Apache 2.0 License.
- Data is not gated behind a vendor export request or a per-event export charge.

## Deployment model

The production design can run on two virtual machines:

- VM one handles ingestion and raw storage: Collector API, NATS JetStream, raw writer, and SeaweedFS.
- VM two handles analytics and access: ClickHouse, dbt runner, Metabase, the Rawbbit MCP server, and supporting services.

The raw Parquet layer is the boundary between the two machines. Everything runs as Docker containers from published images and can be evaluated locally with Docker Compose.

## Ways the studio uses the data

- Metabase dashboards provide shared metrics such as retention, revenue, progression, funnels, cohorts, and session behavior.
- Analysts can query ClickHouse directly with SQL, without sampling or vendor export limits.
- MCP-connected AI agents can translate plain-language questions into read-only analytical queries.
- Optional read-only codebase access can help an AI agent connect behavioral changes to the game systems and mechanics that produced them. Code access is not required for analytics access.

## Product and delivery model

Rawbbit offers two ways to start:

1. Free self-hosting from the public GitHub repository for teams that want to deploy and operate the stack themselves.
2. Setup and maintenance by the Rawbbit team, in the studio's infrastructure or on managed servers, priced by quote.

Rawbbit does not charge per event or per seat. Infrastructure and any selected managed services have their own costs.

## What ships today

- HTTP event ingestion with collector validation.
- NATS JetStream buffering and raw-writer at-least-once delivery.
- Partitioned raw Parquet in SeaweedFS or GCS.
- Scheduled dbt Core loading, testing, reconciliation, and bounded backfills into ClickHouse.
- ClickHouse-backed Metabase dashboards and direct SQL access.
- Read-only MCP tools for analytical queries and operational logs.
- Docker Compose deployment scaffolding and published service images.

## Good fit

- Game studios that want serious analytics without hiring a data engineer.
- Teams where designers, producers, analysts, and leadership need access to the same underlying data.
- Studios that want dashboards for routine metrics and deeper access through SQL or AI agents.
- Teams that care about owning portable raw event history and avoiding per-event SaaS pricing.
- Technical teams that want an open-source stack they can inspect and self-host.

## Poor fit

- Teams looking for a zero-setup, plug-and-play dashboard product.
- Studios where nobody will review the data or own analytics decisions.
- Teams unwilling to own their data or choose a managed operating arrangement.
- Buyers who require native Unity, Unreal, or Godot SDKs before adopting the HTTP endpoint.

## Canonical pages

- Homepage: https://rawbbit.one/
- How it works: https://rawbbit.one/how-it-works
- Pricing: https://rawbbit.one/pricing
- About: https://rawbbit.one/about
- GitHub: https://github.com/mirlan-irokez/rawbbit

## Contact

- Email: team@rawbbit.one