Skip to content

Core Concepts

This page introduces the main building blocks of the Ambient Code Platform. Each concept links to a deeper reference page where available.

A workspace is a project container that groups sessions, integrations, and team members together. Workspaces map to Kubernetes namespaces under the hood, providing resource isolation between teams.

Each workspace has:

  • Members and roles — Control who can create sessions, manage integrations, or administer the workspace.
  • Settings — Default model configuration and API keys.

Learn more in Workspaces.

A session is a single AI agent execution. When you create a session, ACP provisions a containerized environment, clones any requested repositories, and runs the agent with your prompt.

Sessions have a defined lifecycle:

PendingCreatingRunningStoppingStopped / Completed / Failed

While running, you can interact with the agent through a chat interface, observe its progress in real time, and browse output artifacts. Sessions are configurable:

  • Model — Which LLM powers the agent. Available models include Claude Sonnet 4.5, Claude Opus 4.5, Claude Haiku 4.5, and Gemini 2.5 Flash (generally available), plus Claude Opus 4.6, Claude Sonnet 4.6, and Gemini 2.5 Pro (feature-gated, when enabled by your administrator).
  • Temperature — Controls response randomness (default: 0.7).
  • Max tokens — Maximum output tokens per response (default: 4000).
  • Timeout — Maximum execution time in seconds before the session is stopped (default: 300).
  • Repositories — One or more git repos cloned into the session workspace.
  • Workflow — An optional structured template guiding the agent’s approach.

Learn more in Sessions.

Integrations connect ACP to external services so agents can read from and write to the tools your team already uses. Integrations are user-scoped (tied to your SSO identity) and available across all your workspaces.

IntegrationAuth MethodWhat Agents Can Do
GitHubGitHub App or PATClone repos, open PRs, read/comment on issues
GitLabPersonal Access Token (PAT)Clone repos, open merge requests, interact with issues
JiraAPI Token (email + token)Read and update tickets, add comments, transition status
Google DriveOAuthAccess Drive files for context

Once connected, every session can use them.

Learn more in Integrations.

Workflows are structured task templates that guide how an agent approaches a problem. They provide consistent, repeatable processes for common tasks.

Built-in workflows:

  • Bugfix — Diagnose and fix a reported bug, including tests.
  • Triage — Classify, prioritize, and route an issue.
  • Spec-kit — Generate a technical specification from requirements.
  • PRD/RFE — Produce a product requirements document or request for enhancement.

Custom workflows can be loaded from any git repository, letting teams codify their own processes and share them across the organization.

Learn more in Workflows.

Context is the input an agent works with: cloned repositories, linked documents, integration data, and your prompt.

Artifacts are the outputs a session produces: modified files, generated documents, pull requests, and any other files the agent creates during execution. You can browse and download artifacts from the session detail page.

Learn more in Context and Artifacts.

The Model Context Protocol (MCP) lets agents call external tools during a session. MCP tools extend what an agent can do beyond code and text — for example, querying a database, calling an internal API, or running a specialized analysis tool.

MCP tools are configured at the workspace level and made available to all sessions within that workspace.