TestBase Foundations

TestBase combines deterministic multi-agent workflows with infrastructure that feels native to software engineers. Rather than bolting tools onto a single LLM, TestBase coordinates planners, workers, and reviewers running on full computers—locally or inside managed containers—so complex tasks can be delegated with reliable guardrails.

Core building blocks

• Codex Agents — Planner, worker, reviewer, and orchestrator roles execute on real machines with shell, filesystem, and Git access. They inherit Codex CLI capabilities and can add MCP servers for external tools.
• TestBase Containers — Managed, persistent environments running in Google Cloud Run (via TestBase Cloud) that mirror local agents. They sync workspaces to Google Cloud Storage and expose secure APIs for commands, logs, and files.
• MCP (Model Context Protocol) — Standardized interface for attaching auxiliary services (filesystem, search, SaaS APIs). TestBase agents can mix hosted HTTP/SSE servers and locally installed stdio servers.
• Session Artefacts — Every run emits structured logs (docs/sessions/<id>/) including turn transcripts, plan summaries, and reviewer outcomes to make auditing and iteration straightforward.

How the pieces fit together

1. A user or orchestrator agent submits a task.
2. A planner (optional) decomposes the work, emitting a plan artefact.
3. The worker executes changes either on a local workspace or inside a TestBase Container, using MCP servers to reach external systems.
4. A reviewer (optional) validates the diff before completion.
5. Artefacts sync to durable storage (local disk or GCS) for inspection and follow-up.

This orchestration allows teams to mix and match compute locations without changing agent code—computer-agents toggles between local and cloud runners using the same API surface.

Documentation map

• Architecture — Understand the runtime topology and lifecycle across planners, workers, containers, and cloud services.
• CLI Quickstart — Launch your first end-to-end run from the repository and inspect the generated artefacts.
• TestBase Containers — Dive into container lifecycles, persistent workspaces, and the Files API.
• TestBase Agents SDK — Configure agents, connect to cloud containers, and register MCP servers.
• TestBase Cloud — Learn the managed infrastructure, REST API, and operational playbooks.
• Workflow Guides — Walkthroughs that combine local and cloud execution, automation hooks, and tool integrations.
• Reference — Configuration cheat sheets, glossary, and links back to source-of-truth docs in the repo.

Move at your own pace—the sections above are designed to be browsed independently once you’re familiar with the foundations.