Observal
Observal is an Observability and Evaluation platform for human-in-the-loop agents
Last verified:
What is Observal?
Observal is an open-source registry and observability platform for AI coding agents. It allows teams to browse, install, evaluate, and trace agents across Claude Code, Cursor, Kiro, Gemini CLI, and more. Think of it as Docker Hub but for AI coding agents — every agent bundles its MCP servers, skills, hooks, prompts, and sandboxes into a single portable YAML package that installs with one command.
The platform provides three core capabilities: an agent registry for browsing and publishing complete agent configurations, a telemetry pipeline that traces every tool call, token, and session (streaming into ClickHouse), and an evaluation engine that scores agent sessions on goal completion, tool efficiency, factual grounding, thought process, and adversarial robustness. The dashboard shows live agent scores, recent sessions, top downloads, error logs with drill-through, and AI-generated insight reports on usage patterns.
Observal is designed for AI engineers who want to know which agents actually deliver results, platform teams running agents across an organization who need visibility and governance, agent authors who want to share their work and see real-world performance, and operators who want the entire stack self-hosted with Docker Compose and no SaaS egress costs.
Observal pricing
Pricing model: Freemium
Observal is free and open source under the Apache 2.0 license. You can self-host the entire platform for free with no cost. There is an enterprise edition available that includes SSO, SCIM, and audit logging features, but everything in the core documentation works without the enterprise edition. No paid tiers or usage-based pricing is mentioned for the open-source self-hosted version.
Observal pros
- Open source under Apache 2.0 license
- Self-hosted with Docker Compose, no SaaS required
- One-command installation: curl -fsSL https://observal.io/install | bash
- Portable YAML agent format works across multiple IDEs
- Bundles MCP servers, skills, hooks, prompts, and sandboxes together
- Traces every tool call, token usage, and session
- Built-in evaluation engine with structured rubrics
- Scores on goal completion, tool efficiency, factual grounding, thought process, and adversarial robustness
- ClickHouse backend for fast telemetry queries
- 零 configuration needed for MCP servers
- Admin approve/reject workflow for component submissions
- AI-generated insight reports on usage patterns
- Classified error logs with drill-through to triggering sessions
- Supports Claude Code fully with skills and hook bridge
- Supports Kiro CLI fully with superpowers and steering files
- Executive dashboard with cost intelligence and velocity metrics
- Department breakdowns for organization-wide AI adoption visibility
Observal cons
- Self-hosting requires your own infrastructure and ClickHouse
- No managed SaaS option mentioned
- VS Code support is limited to MCP servers and rules only
- Copilot CLI support is limited
- Codex CLI support is limited to rules only
- OpenCode support limited to JS plugin hook bridge
- Enterprise edition (SSO, SCIM, audit logging) is separate from open source
- Requires Docker Compose knowledge to deploy
Frequently asked questions about Observal
What is Observal?
Observal is a self-hosted AI agent registry with built-in observability. Think Docker Hub, but for AI coding agents. It provides a registry to browse, publish, and install complete agent configurations; a telemetry pipeline that traces every tool call into spans, traces, and sessions streaming into ClickHouse; and an evaluation engine that scores agent sessions on multiple criteria.
Which IDEs and tools does Observal support?
Claude Code and Kiro CLI are fully supported with skills, hook bridge, MCP servers, rules, and OTLP telemetry. Gemini CLI and Cursor are tested with hook bridge, MCP servers, and rules. VS Code, Copilot CLI, Codex CLI, and OpenCode have limited support for MCP servers, rules, and hook bridges.
Is Observal free?
Yes. Observal is open source under the Apache 2.0 license. You can self-host the whole thing for free. There is an enterprise edition with SSO, SCIM, and audit logging, but everything covered in the documentation works without it.
How do I install Observal?
Install with one command: curl -fsSL https://observal.io/install | bash. The platform runs via Docker Compose on your own infrastructure with no SaaS required.
What does the evaluation engine score?
The evaluation engine scores agent sessions on goal completion, tool efficiency, factual grounding, thought process, and adversarial robustness. This lets you compare agent versions and see what actually works in production.
Who is Observal for?
Observal is for AI engineers who want to know which agents deliver real results, platform teams running agents across a team or org who need visibility and governance, agent authors who want to share their work and see performance on real workflows, and operators who want the entire stack self-hosted with no SaaS egress.
What data does Observal collect?
Every tool call becomes a span, spans group into traces, and traces form sessions. The platform captures models used, token counts, tool breakdowns, and 16 turns of interaction data. Everything streams into ClickHouse without requiring changes to your MCP servers.
Can I publish my own agents to Observal?
Yes. Agent authors can publish their complete agent configurations as portable YAML files. Each agent bundles MCP servers, skills, hooks, prompts, and sandboxes. There is an admin approve/reject workflow for submitted components in the review queue.
What license is Observal under?
Observal is open source under the Apache 2.0 license. The website also mentions AGPL-3.0 in some places, indicating the project may use different licenses for different components.
What does the dashboard show?
The dashboard displays live agent scores, recent sessions, and top downloads. It also shows error logs with classified errors and drill-through to triggering sessions, AI-generated insight reports analyzing usage patterns, and executive dashboard metrics including cost intelligence, velocity metrics, and department breakdowns.