Search documentation

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

User manual · Start

See Awaken solve a real Agent job

What this page covers

Choose one of six Agent stories and see the job, product evidence, human control point, and finished result before you watch.

These six videos answer different product questions. Pick the question that matches your work rather than watching them in order.

The demonstrations use controlled test inputs. The SDK calls, Sessions, Console state, tool execution, approval boundaries, restarts, and artifacts come from a running release build. Some stories use a fixed Agent response so the demonstration does not depend on a model deciding whether to call a tool. The video states this when it applies. These are product demonstrations, not customer deployments.

Choose a story

StoryJob shownEvidence you seeWhat you learn
Evidence to decisionRead a fixed evidence file and return a reviewable decisionRead-only input, tool trace, grounded HOLD, next actions, and retained approval authorityAwaken can deliver a traceable result instead of an answer with no source
Build entirely in ConsoleConfigure an API compatibility reviewer, test a breaking contract diff, then publish the reviewed AgentModel, instructions, tools, permissions, resources, context, both protocol views, the compatibility result, and publication diffConsequential Agent controls are visible and reviewable in the UI before the exact revision becomes reusable
Connect with the Anthropic SDKStart work from an Anthropic Managed Agents client and inspect the same Session in ConsoleOfficial SDK request, committed events, one Session identity, and Console readbackAn existing backend can connect without creating a second history model
Human-controlled actionInspect a pinned repository source and prepare a protected review artifactSource revision and hash, completed read, pending write, human approval, and downloadable artifactAn Agent can reach a consequential tool while a person keeps authority
Survive restartRestart Awaken while an approved work item waits, then continue itOriginal Session, pending approval, new AllInOne process, one resumed write, and attached artifactAccepted work and its approval boundary survive routine service recovery
Scheduled operationTurn a repository snapshot into a recurring exception reportRead-only snapshot, Deployment, Run once, separate result Session, and human next actionA tested Agent can become repeatable work without becoming an opaque cron task

Together, the videos cover the main journey: build an Agent, connect an application, run real work, keep authority over protected actions, recover committed work, and schedule a repeatable operation.

What makes a video publishable

flowchart LR
  J[A concrete technical job] --> C[Real Console or API operation]
  C --> E[Visible product evidence]
  E --> R[Useful result]
  R --> H[Human review or next action]
  H --> P{Did every checkpoint pass?}
  P -->|yes| V[Publish this recording]
  P -->|no| D[Keep diagnostics and publish nothing]

A public story has a concrete job, a useful result, a visible human-control point, and a clear next action. It does not need a fictional customer, job title, deadline, or emergency to make the work sound important.

The recording is current only when a fresh release run reaches every checkpoint. If a dependency, product action, or assertion fails, the run keeps its diagnostics and rejects the recording. An older successful file cannot stand in for the failed run.

The source evidence for this page establishes the product mechanisms, not whether a particular MP4 passed. Recording commands stay with the release assets and are not published on this page. When a checkpoint fails, that run does not establish the claimed effect and publishes no current recording.

Check what the video proves

Before you rely on a demonstration, confirm four things on screen:

  1. The opening names the job and the result you will receive.
  2. Each claim points to visible Console or API evidence.
  3. The result closes the loop with an artifact, decision, or next action.
  4. The recording identifies the Awaken revision used for verification.

Use Get started with Awaken to reproduce the basic journey. Continue with build an Agent, connect an application, or manage a Session for the workflow shown in the video you selected.

Review the implementation in the Awaken repository. If the evidence is useful to your evaluation, star the repository so other teams can find it.