Search documentation

Browse Awaken Agents docs
Docs/Awaken Agentsv1.0.0-dev/User manual/StartStart with Awaken
Note·You're reading pre-release documentation (v1.0.0-dev). Interfaces and behavior may change before a stable release.

User manual · Start

Start with Awaken

What this page covers

Decide whether Awaken should own your Agent execution, then run one durable Session and choose the next integration path.

Awaken is a self-hosted platform for running Agents as durable Sessions. An application publishes Agent behavior, sends work through a supported protocol, and reads committed events from the same Session after a wait, retry, Worker change, or process restart.

Start here when the application should not own its own Agent loop, Tool execution, placement, persistence, and recovery.

Choose the smallest boundary

NeedUseIt owns
One stateless model responseA model APIrequest and response
Durable Agent execution behind an application APIAwaken AgentsAgent publication, Session, policy, dispatch, execution placement, and committed events
Work ownership, dependencies, human decisions, and acceptance around Agent executionAwaken Workforcethe accountable work record

Keep these boundaries separate. A Session records Agent execution; it does not become the business acceptance record. A frontend protocol changes how input and events travel; it does not create another Agent implementation.

The application contract

An application works with four objects:

ObjectStable responsibility
AgentVersioned instructions, model choices, Tools, MCP servers, Skills, delegation, and limits
EnvironmentExecution placement, isolation, resources, and network constraints
SessionOne durable Agent instance bound to an Agent publication and Environment
EventInput, progress, Tool interaction, usage, status, and committed output

Workers, Sandboxes, stores, and the Runtime implement this contract. They are deployment and execution components, not additional application-facing records.

flowchart LR
  C[Client protocols] --> S[Session application]
  A[Agent configuration] --> P[Immutable publication]
  P --> S
  S --> D[Run ingress]
  D --> W[Worker and Sandbox]
  W --> R[Runtime]
  R --> X[Models · Tools · MCP · external Agents]
  R --> K[Commit boundary]
  K --> F[Committed Session facts]
  F --> S
  S --> C

Managed Agents, AI SDK, AG-UI, A2A, and MCP all enter this one contract. Choose between them in the protocol connection matrix.

What happens in one Session turn

sequenceDiagram
  participant App as Application
  participant Agents as Agents control plane
  participant Ingress as Run ingress
  participant Worker
  participant Runtime
  App->>Agents: Create Session and append input
  Agents->>Ingress: Admit one Run
  Ingress->>Worker: Claim a fenced attempt
  Worker->>Runtime: Execute the pinned Agent snapshot
  Runtime-->>Agents: Stream interaction events
  Runtime->>Agents: Commit step and terminal facts
  Agents-->>App: Return replayable events and idle status

Queueing, bounded retry, waiting for approval, and lease recovery stay inside their owning mechanisms. External action begins only when the Session exposes a terminal or attention result. Production reliability explains those results and the corresponding action.

Run the first path

Follow Run your first Awaken Session. It starts an AllInOne process, connects one provider, publishes one Agent, sends input with the official Anthropic SDK, and reopens the committed Session after a restart.

The path is complete when the application receives Agent output, observes session.status_idle, saves the Session id, and can read the same committed events again.

Continue from the next task

Next taskContinue with
Connect a frontend, backend, or compatible AgentConnect a published Agent
Choose local, durable, or multi-Worker deploymentSelf-host Agents
Understand control, persistence, execution, and isolationAgents architecture
Change prompts, models, Tools, Skills, or limits without rebuildingConfigure Agent behavior
Add a Rust Tool, provider adapter, Plugin, or execution behaviorExtend Awaken Agents internals

Awaken Agents is open source, and its first stable release is coming soon. Its interfaces and behavior may still change before that release. A hosted service, support commitment, security certification, or independently measured production result requires a separate published statement.