Observal packages internal agents for Claude Code and Cursor

Observal is pitching itself as a self-hosted control plane for internal AI components: a place to publish, discover, install, and analyze the agents, MCP servers, prompts, hooks, and sandboxes that teams increasingly build for coding tools.

The project argues that many organizations now build internal AI tooling, but store it in scattered GitHub repositories with limited documentation and little feedback on whether those components actually help developers. Observal’s answer is a centralized registry plus session-level analytics intended to make those tools reusable and easier to improve.

Supported harnesses include Claude Code, Kiro, Cursor, Pi, Copilot for both CLI and VS Code Extension, Codex, OpenCode, and Antigravity CLI. The core idea is to define an agent once, then let Observal generate the right configuration for each supported environment.

A deployable setup has two parts:

• a self-hosted server with API, web UI, and databases
• a CLI installed on each developer machine

The one-line server install requires Docker Engine 24.0 or newer with Compose v2. The stack includes PostgreSQL, ClickHouse, Redis, a worker, load balancer, Prometheus, and Grafana. The CLI can be installed as a standalone binary or through Python 3.11+ using uv or pipx.

Recommended reading

South Africa mining automation may widen inequality

Beyond distribution, Observal emphasizes observability. It can capture sessions automatically after authentication, then surface reports on agent adoption, prompts, tools, and workflows. It also offers session replay with user prompts, thinking blocks, assistant responses, and tool calls, aimed at debugging, audits, and governance.

The open-source release is under Apache License 2.0 and includes features such as audit logs, SAML SSO, SCIM provisioning, and an executive dashboard in the Apache-2.0 distribution. Under the hood, the stack uses Vite 6, React 19, Python 3.11+, FastAPI, Strawberry GraphQL, PostgreSQL 16, ClickHouse, Redis + arq, and supports deployment through Docker Compose or Kubernetes with Helm.

For teams already standardizing around coding agents, the pitch is simple: one registry, one install flow, and a clearer record of what those internal tools are actually doing.