Create and connect

Version Workspace

The version workspace is the control plane for one version. Here you define runtime capabilities, presentation information, and pricing; start a local build; and ultimately publish. The remote Program files live under /oasn in the development Workspace. Do not confuse the two.

Overall layout of the OpenAgent version workspace
Version workspace overview. The page brings together Program, Agent Card, Pricing, Save Draft, Build With Local Agent, and Publish. The screenshot uses sanitized test data.

Page areas and responsibilities

AreaWhat you do hereCompletion signal
ProgramView development VM/Chat entry points, select models under Models, and choose CPU/Memory under ResourcesRequired models and CPU/Memory are selected; the Program has been tested in the Workspace
Agent CardSet the avatar, name, description, category, and MOMAll required fields pass page validation, and a first-time caller can understand the MOM
PricingSet the cost-plus gross margin and Minimum Starting BalanceThe gross margin is valid, and the minimum starting balance is expressed in Credits and is at least 0
Save DraftSave the current form configurationThe page no longer shows unsaved state or field errors
Build With Local AgentGenerate a trusted build prompt bound to this versionThe complete prompt is passed unchanged to the local AI
PublishConfirm and start version publishingThe publish job succeeds; Go Online is still a separate action

Program: declare runtime requirements first

Complete at least the following configuration before treating the version as a publish candidate:

  • LLM: currently a required model capability. Select the model capabilities the Agent actually calls.
  • Optional understanding capabilities: select image, video, audio, PDF, image understanding, or similar capabilities only if the Program truly processes those inputs.
  • Resources: the current dialog provides only CPU and Memory, and both must be set. Size them for normal request peaks, not merely for whether the process can start.
  • Plugin/MCP/A2A: these are not configured in the Resources dialog. Implement them in the Workspace as described in the Program section and validate them with real calls.
Models configuration dialog in the version workspace
Models configuration. LLM is required. Select other capabilities according to the Program's real inputs and calls, and do not declare capabilities you have not tested.

A page declaration does not prove the implementation works

After selecting models and CPU/Memory, you must still validate real business requests in the development Workspace. If the Program uses a Plugin, MCP, or A2A, separately verify its runtime, business state, and final output.

Agent Card: tell callers how to use the Agent

The current workspace lets you edit the avatar, name, description, one category, and MOM. Publish readiness requires an HTTPS avatar. Uploads support PNG, JPEG, or WebP up to 5 MiB; a 1:1 image is recommended. The MOM must be a UTF-8 Markdown file no larger than 1 MiB and must contain explicit Use when, Inputs, and Outputs sections.

A background image, tags, and website URL are not part of the current workspace flow. Do not search for controls that exist only in older documentation.

Pricing: configure only the currently public model

The current workspace uses cost-plus pricing. Gross margin must be an integer percentage from 0 to 100. Minimum Starting Balance (minimum_starting_balance_credits) is denominated in Credits and must be at least 0. A decimal gross margin may be saved in a draft, but Publish rejects it with PRICING_MAPPING_INVALID; do not rely on silent rounding. Before publishing, use representative requests to confirm that the resulting cost matches your expectation.

When to save the draft

  1. After creation, enter the version number and minimum runtime configuration, then click Save Draft.
  2. After one Workspace development iteration, complete models, resources, Card, and Pricing, then save again.
  3. Before publishing, reopen the page and confirm that the loaded values match your expectation. Do not treat unsaved browser state as server-side fact.

Current implementation boundaries

Available now

  • Save version configuration and start publishing on the website; after success, explicitly Go Online or use Take Offline when needed.
  • Copy a trusted prompt that contains the fixed Skill URL and bound IDs.
  • With CLI 0.7.3, get, set, or refresh WebUI ports and use allow-listed portal operations for draft/model/resource configuration and publishing; mutating operations require explicit authorization and --confirm.

Not available now

  • A public Workspace close button or CLI close command.
  • A WebUI port field on the website form. Use CLI 0.7.3 version webui-ports get, set, or refresh instead.
  • A generic CLI file synchronization command. Synchronize reviewed files through the explicit staged SSH/SCP procedure.

Next: Hand the version safely to a local AI for building.

Search documentation

Suggested searches