Core concepts
The user-facing vocabulary used across the app, SDKs, and docs.
Agent
An agent is the thing the user chats with in Pacerelle. Behind it, you run a local process with an agent id and token.
Examples:
- a local Ollama assistant
- a Python script that can read a local folder
- an MCP bridge for Claude Desktop or Cursor
- a workflow runner that executes release tasks
Conversation
A conversation is the chat between one user and one agent. Messages, files, widget requests, and widget responses all appear inside a conversation.
Trusted browser
A trusted browser is a browser profile that has signed in and prepared local secure state for the account. Clearing browser data can remove that local trust state.
Agent id
The agent id identifies the agent record in Pacerelle. Copy it from the New agent confirmation panel into your local .env file.
Agent token
The agent token lets your local runtime connect as the agent. It is shown once when the agent is created or rotated. Keep it secret.
SDK
The SDK is the easiest way to build a local agent runtime. It receives user messages, exposes attachments and widget responses, and sends replies back to the conversation.
MCP server
The MCP server is a packaged local bridge for MCP-compatible hosts such as Claude Desktop, Cursor, and Zed. Use it when you want Pacerelle integration without writing your own runtime.
Widgets
Widgets are structured UI prompts inside a conversation. Use them when the agent needs approval, a file, a choice, a form, or progress feedback.
End-to-end encryption
Pacerelle is designed so message content is encrypted before it leaves the browser or local agent. As an agent builder, the main rule is simple: avoid logging decrypted user content unless the user explicitly expects it.