Server

The server receives messages from clients: the TUI, the web renderer, your code and other agents. These settings control where it listens and what it accepts.

Example

[server]
bind = "0.0.0.0:8080"
base_url = "https://agent.example.com"
cors_origins = ["https://app.example.com"]

[server] settings

Setting	Default
base_url	none: you must set it
bind	"127.0.0.1:8080"
cors_origins	none
ui_dir	none
max_request_bytes	1048576 (1 MiB)
rate_limit_per_minute	120
bridge_interval	"PT0.25S" (250 milliseconds)
delegators	none

base_url

The address that clients use to reach the server. The server puts it in its agent card, the public description that clients read first.

• Default: none. You must set it.
• Example: "https://agent.example.com" behind a gateway, or "http://127.0.0.1:8080" on your computer.
• Other agents send updates to this address. If they cannot reach it, set updates = "follow" for those agents.
• A path in the address, as in "https://example.com/agents/billing", is where your app serves the agent. See Mount agents on your app. tiny-harness serve also uses it, for a gateway that forwards that path.

bind

The network address and port where the server listens.

• Default: "127.0.0.1:8080". Only programs on the same computer can connect.
• Set it when: the server runs in a container or behind a gateway on another host. "0.0.0.0:8080" listens on all networks. Then a gateway must protect it. See Deploy.

cors_origins

The websites whose web pages can connect to the server from a browser.

• Default: none. A web page from another website cannot connect.
• Set it when: you use a web renderer from another website. For the web renderer on this documentation site, add "https://madarauchiha-314.github.io".
• Example: cors_origins = ["https://app.example.com"]

ui_dir

A folder with a built web renderer. The server shows it at /ui/.

• Default: none.
• Example: ui_dir = "renderers/web/dist" in a copy of the repository.

max_request_bytes

The largest request the server accepts, in bytes. A larger request gets the error 413.

• Default: 1048576 (1 MiB)

rate_limit_per_minute

The number of requests one client can send in one minute. More requests get the error 429.

• Default: 120

bridge_interval

How often the server checks Temporal for new events to stream to clients.

• Default: "PT0.25S" (250 milliseconds)
• Set it when: you want faster updates (a smaller value) or less load on Temporal (a larger value).

delegators

The other agents that can give this agent a full task. A full task sets how the task ends and which model it uses.

• Default: none. Only this agent itself can give itself a full task. It does this when it runs a step in a sub-task on itself.
• Set it when: another tiny-harness agent gives work to this one, and you trust it. Then its sub-tasks end by themselves when they are done. Without it, the other agent still gives work to this one, but its sub-tasks stay open after they answer.
• Example: delegators = ["front-desk"]. Use the participant id that the other agent sends.
• Any other sender that tries to set these values gets an error, and tiny-harness does not create the task.