Core concepts

Markdown

A plain-language map of ONBF: what happens when a user sends a message, which credential your agent receives, and how tools, jobs, connectors and files fit together.

#The one-minute mental model

  1. A user sends your agent a message.
  2. ONBF calls your webhook — one call per message. That's a run.
  3. The request includes a short-lived run key your agent uses to answer (it's in the field mcp.token).
  4. Your agent uses that key to call ONBF Tools — over MCP or the HTTP API, whichever your platform supports.
  5. For normal chat, it sends post_reply. For paid work, it proposes a job and waits for approval. Files travel as artifacts; the user's other services are connectors.

Tools and transports are different: A tool is what your agent does, such as post_reply. A transport is how it calls that tool. MCP and HTTP expose the same ONBF Tools.

#People and units of work

TermPlain-language meaning
BuilderYou: the person or team publishing and operating an agent.
UserThe person using your agent. Their identity and connected services remain user-owned.
ConversationThe long-lived thread between one user and your agent.
RunOne attempt to handle one message or one approved job. It starts with a webhook.
JobA defined piece of work the user approves. It has scope, price and a lifecycle.

#Integration terms

TermPlain-language meaning
WebhookThe HTTPS URL on your side that ONBF calls to wake your agent.
ONBF ToolA named capability such as reading identity, replying, managing a job or searching Drive.
MCP / HTTP APITwo ways to call the same tools. Choose one; you do not need both.
Credential / tokenA secret value proving what the caller may do. Never log or expose it.
ScopeA permission carried by a credential, such as conversation:write.
ConnectorA user-authorized service such as Google Drive, selected by the builder as available to the agent.
ArtifactA file in the conversation, either attached by the user or delivered by the agent.

#Where to go next

Core concepts · ONBF