Skip to main content
Create an agent with New in the left column, or describe one to Omni and review what it builds. All settings below are in Agent settings, the gear in the conversation header. Edits save to the live agent on their own.
Agent settings panel showing Agent, Skills, Knowledge, Access, Schedule, Channel, Prompt and Advanced rows beside the composer with Claude Code, Opus 5.5

Agent settings: skills, knowledge, access, schedule, channels and prompt on one card; harness, model and effort are set from the composer. Shown with sample data.

Instructions

PROMPT opens the prompt editor:
  • Static: freeform instructions. Generate drafts them from a short description.
  • Dynamic: a Python function, xpander_dynamic_prompt(), that runs on every turn and adds its return value before or after the static prompt. Workspace secrets are available as os.environ.
Omni reads the agent’s description to route requests to it.
Prompt editor in Static mode and Edit, with Generate, Cancel, Save and Copy buttons and a short helpdesk prompt typed in

The Static prompt editor in Edit mode: a freeform instructions box with Generate, Save and Copy. Shown with sample data.

Harness and model

The composer’s Harness and model picker sets Claude Code, Codex or OpenCode and the model, from the providers your admins enabled under Platform > AI governance > Models. The picker lists models only, not aliases. Any harness can run a model from any enabled provider: the AI Gateway translates each call. The effort chip next to it (Default, Low, Medium, High or X-High) sets how hard the model reasons, on every harness. Default is the model’s own level. A level the model does not take is greyed out with the reason, for example This model stops at High. The pick applies to this conversation from the next turn, not to the agent. The chip shows unless your user group turns off Effort level.
Claude Code can run a GPT or Gemini model, Codex can run a Claude model, and all three can run your own provider’s model. On Claude Code and Codex, the picker lists a model from another vendor only when that pair passed xpander’s test on that harness. OpenCode lists every catalog model except a pair the test showed as broken. Your own provider’s model is listed after Test model passes for that harness.A model you pick inside a conversation applies from the next turn. If you can edit the agent, a model from the CLI’s own vendor, or from a custom provider, is also saved as the agent’s default model for that CLI. OpenCode saves a model from any provider. If you cannot edit the agent, the picker shows Saved for you and the agent’s default does not change. A pick that does not become the default, such as another vendor’s model on Claude Code or Codex, stays in your own settings for this agent, and your new conversations on that CLI start on it.When the default changes, the next turn runs on:To move an agent off a model, change the default, then pick a model in each conversation that has its own.An agent-level credential in the Vault can set the agent’s own provider key or endpoint. It outranks the organization’s key. Model tokens on your own key are measured but not charged in credits. On xpander cloud, the run pays for session hours instead (Budgets and usage).

Skills

Add skill offers the skills your organization made available. Per skill, you choose which actions are enabled and which wait for approval. See Skills and Add a skill.

Approvals

A skill’s actions, an API connector’s approval policy and Gated commands on the agent can make an action wait for a named person. See Approvals and Permissions.

Secrets

A skill’s credential is entered when you connect the skill, and the agent never sees it. The agent’s own secrets are under Advanced > Vault secrets. See Credentials.

Output format

Output format is Default, Markdown or Structured output (a JSON Schema). For answers that must always have the same sections, write them into the prompt. See API and structured output.

Runtime environment and built-in tools

Advanced > Runtime environment attaches the agent’s computer: the CLIs, packages, config and context it finds on every run. See Runtime environments. Harness built-in tools and xpander built-in tools, under Advanced, can be turned off so the agent works only through what you attached. See Built-in tools.

Budgets and limits

Advanced holds a Per-agent budget and a Per-turn budget. See Budgets and usage.

Rehearse with mock data

Each connection has a Live / Mock switch. In Mock, xpander returns generated, cached data in the operation’s response shape. Nothing reaches the vendor and no credential is used. The run still uses the model and is recorded. There is no whole-agent dry run.

Saving changes

Edits save to the live agent within a second or two. There is no draft, staging copy or publish step: the next turn, API call or scheduled run uses the new settings. The prompt editor saves when you select Save. Sharing and editing rights: Permissions.

Roll back to an earlier version

Select the version chip next to the agent’s name in Agent settings, for example v7, pick a version and select Publish to live. Or ask Omni in chat, for example “roll back the support agent”.
Version history keeps about one version per editing session: edits close together update the same version, and a new version starts after 30 minutes without edits. The version chip shows only when your user group has Version history on (Platform > Organization > Members & groups > Groups, a group, Customize features).
  1. Select the version chip. Version history opens with this agent’s versions.
  2. Select a version. Inspect shows how its settings differ from the agent’s current state; skill and knowledge base changes are not listed. Compare puts two versions side by side.
  3. Select Publish to live. That version’s settings and instructions go live at once as a new version number, so the history keeps the version you left. Its skills, connected actions and knowledge bases are not restored: the agent keeps the ones it has now.
Omni restores the whole version before the current one, or the version number you name, skills and knowledge bases included.

Next steps

Add a skill to an agent

Steps per skill kind.

Set up a runtime environment

CLIs and context.

Decide who can use the agent

Access and sharing.

Get structured answers from the API

The same shape every time.