Agents

Agents are memory identities for AI workflows.

Use them when a workflow needs its own memory boundary instead of relying on a person, one chat, or the whole organization.

Agent modes

Mode	Use it for	Memory behavior	Price
Durable	Recurring responsibilities and automations	Memory carries across future runs	$10 per agent per month
Ephemeral	One-time executions, tests, previews, and temporary workflows	Memory is limited to the execution or session boundary	$10 one time

When to use an agent

Agents are useful when you want a workflow to keep its own context, such as:

• release-readiness tracking;
• customer-specific implementation work;
• support triage for one account; or
• a developer workflow with its own history and constraints.

Runtime model

Agents use the same broad shape as OpenAI chat completions. In the current app, agent-scoped inference is exposed through the inference gateway.

You can identify the agent either by route or by header, depending on the client flow you are using.

curl https://your-org-slug.achiral.ai/v1/chat/completions \
  -H "Authorization: Bearer acm_..." \
  -H "Content-Type: application/json" \
  -H "X-Achiral-Agent: release-readiness-agent" \
  -d '{
    "model": "chiro",
    "messages": [
      {
        "role": "user",
        "content": "Prepare the next rollout review with prior migration context."
      }
    ],
    "temperature": 0.2
  }'

For the route-based contract, see API reference.

Memory boundaries

Agent retrieval is narrowed to the requesting organization, token scope, and the selected agent boundary.

Shared organization context can still matter, but agent-local continuity is the default.

Pricing notes

• Durable agents are $10 per agent per month.
• Ephemeral execution agents are $10 one time.
• Founder workspaces can create one draft preview.
• Activation requires a Team plan or Custom contract because Memory Agents are a paid add-on.

Next

• Quick start
• Runtime and scopes
• Memory API

For deeper operational detail, see Runtime and scopes.