Dagger

Dagger is a programmable CI/CD engine that runs pipelines as containerized code rather than YAML. Pipelines are defined using Go, Python, TypeScript, or PHP SDKs and executed by the Dagger engine, which manages container isolation, caching, and parallelism. The engine exposes a GraphQL API that both the SDKs and external callers can use to orchestrate pipeline steps. Agent relevance is high: agents can author Dagger pipelines as code, trigger them programmatically, and inspect execution state through the GraphQL API. Dagger Cloud adds distributed caching and execution traces. Open-source (Apache 2.0). Confidence is docs-derived.

Dagger's primary interface is through its SDKs (Go, Python, TypeScript, PHP), which generate strongly-typed GraphQL calls to the Dagger engine. The engine's GraphQL API is comprehensive: container operations (from, exec, withFile, withMountedDirectory), file system manipulation, service execution, and module composition. The Dagger Module system allows packaging reusable pipeline components. The CLI (`dagger call`) provides a REPL-like interface. Direct GraphQL access is possible but the SDK path is recommended for type safety. The API surface is expansive and well-designed for programmatic CI/CD.

Dagger pipelines return typed errors through the SDK: failed exec steps surface exit codes, stdout/stderr, and container state. The GraphQL API returns standard GraphQL error objects with message and locations. Container exec failures are non-destructive (the pipeline errors, underlying containers are cleaned up). Dagger's caching model means retries often skip successful steps. The engine is designed for reliability; partial pipeline failures are isolated to affected steps. Error messages include container output, making debugging practical.

Dagger Cloud uses authentication tokens (DAGGER_CLOUD_TOKEN) for trace uploads and distributed cache access. The local Dagger engine (running as a daemon) requires no auth for localhost calls. SDK calls are authenticated via environment variables when Dagger Cloud features are used. The GraphQL API itself on a local engine has no auth. For remote Dagger engine deployments, network-level controls apply. The Cloud offering uses standard bearer token auth.

Documentation at docs.dagger.io covers getting started with each supported language, the Dagger API reference, module system, and Dagger Cloud. SDK-specific guides are detailed with practical CI/CD examples. The GraphQL schema is introspectable. Dagger Cloud documentation covers traces and caching. Community support via Discord. Documentation quality is high and actively maintained. The project is young but maturing; some advanced module composition patterns are still being documented.