Your first useful agent, step by step.

This guide describes the current private preview. Availability labels distinguish working features from the roadmap.

Workflow quickstarts

01 · Create your workspace

Select Continue securely. In the private preview, each authenticated identity has an isolated personal workspace.

02 · Give your agent a focused objective

Open Agents → Create agent. Choose a starting template, name your agent, and describe the outcome you want. Save changes in the builder before testing.

03 · Add approved knowledge

In Knowledge, upload UTF-8 .txt or .md files, up to 100 KB each. The current search matches keywords and returns exact passages. Avoid uploading secrets or sensitive personal information.

04 · Configure Claude server-side

The operator configures ANTHROPIC_API_KEY and ANTHROPIC_MODEL as server-only secrets. Verify connectivity in Settings → Run Provider Health Check.

05 · Run grounded generation

Choose AI generation in the playground. Claude processes your question together with agent instructions and approved source excerpts, returning grounded answers with citations.

06 · Inspect sources and token usage

Open Runs to inspect inputs, outputs, observable tool events, latency, input tokens, and output tokens. Refine instructions based on factual evidence.

07 · Handle an approval checkpoint

When an agent proposes a consequential action (e.g. data update or refund request), the run halts in awaiting_approval state. Open Approvals to approve or reject.

08 · Publish a private version

Publish version saves an immutable private configuration snapshot. Use Versions to restore a previous configuration to your draft at any time.

09 · Delete workspace data

In Settings, submit a workspace deletion request by confirming DELETE MY WORKSPACE. The workspace freezes immediately and queues for operator removal.

10 · Understand current preview limits

Free preview caps: 10 agents, 10 sources per agent, 50 daily test runs. Read-only tool execution. No external writes or automated webhooks are enabled.

Troubleshooting guide

Missing Anthropic key

ANTHROPIC_API_KEY is not set in server secrets. The model playground falls back to knowledge retrieval tests.

Provider authentication error

The configured API key was rejected by Anthropic. Check provider key validity in Cloudflare dashboard.

Rate limit reached

The workspace exceeded 50 daily test runs or provider API limits. Wait for the daily quota reset or check provider console.

Request timeout

Generation took longer than ANTHROPIC_TIMEOUT_MS (30s). Shorten agent instructions or reduce uploaded document size.

Invalid structured output

The model generated JSON that failed schema validation. Check agent instructions to avoid conflicting format prompts.

No knowledge match found

Keywords in the question did not match uploaded documents. Use more specific keywords or enrich knowledge documents.

Source validation failure

The model cited a source ID not present in retrieved excerpts. The server automatically flags the response as ungrounded.

Run awaiting approval

The execution generated a consequential action proposal. Navigate to the Approvals tab to approve or reject.