API Reference wardrik v#0.1.0-rc.1

Agents, governed: the agents your app runs, the policies they're held to, and the operations people and programs reach them through.

Tools that let an agent work with agents: agents/* for the other agents in its org, and self/* for itself.

Follows the run that answers a message, through the agent's log.

Decides whether a scope may take an action on a resource.

Maps each WardrikProtocol operation to its handler and its compiled input schema.

A budget all of an org's agents share. Set ceilings in the org's Wardrik.OrgSettings

The interrupt a model call waits on once its org's agents reach a ceiling's limit.

Installs Wardrik's chat apps into people's workspaces, from your app's own pages.

The processes Wardrik supervises, in start order

Middleware that summarises an agent's older history when a turn starts.

Agents defined outside the API, such as agent files a daemon reads.

Stores each agent's definition as it was given.

Sends an agent's answer back to the agent that asked with agent_ask.

Decides which URLs users may point Wardrik at, such as when an org adds an MCP server.

Serves Wardrik over HTTP. See WardrikHTTP for the routes.

Turns view results into JSON-ready operation results.

Finds the API keys an agent's code calls services with. Implements WardrikCore.Keys.

Follows agents' events into your app.

Gives an agent the tools of the MCP servers it names. Implements WardrikCore.Capability.

Tells people when an agent parks, fails or finishes while nobody is watching.

Follows every agent's log for Wardrik.Notify.Server.

Posts notices to a room in one of Wardrik's channels, with Wardrik.Channel.notify/4.

Sends notices. One runs per cluster.

Posts notices as JSON to a URL.

Operations on one agent: messaging it, deciding what it waits on, and reading it.

Operations over the agents that resource patterns cover: listing them and their events.

Records changes made through Wardrik in the audit, with who made them.

Passes chat apps' requests to Wardrik's channels.

Operations on the deployment: the audit, nodes, moving agents and reloading.

Operations on an agent's lifecycle: defining, starting, stopping, clearing and forgetting it, and killing its tool calls.

Operations on what an org or person holds for their agents: connected accounts, secrets and MCP servers.

Finds the agents operations act on, waking them from their definitions if needed.

Reads agents' status and logs for operations.

Operations on the deployment's API tokens: creating, listing and revoking.

Operations on an agent's triggers: setting, removing, and delivering webhooks.

Limits operations over many agents to what the scope may see.

An org's settings, as Wardrik.Orgs.settings/2 returns them.

Gives each org its own settings, alongside the deployment's. See Wardrik.OrgSettings.

Runs operations the same way for every surface

Builds %Wardrik{} from Wardrik.start_link/1's options, and each agent's options.

Resumes calls that wait on a sign-in, once the person signs in.

Puts defined agents to sleep after their idle timeout. An agent wakes when something asks it.

Counts what each agent spends per UTC day, in tokens and, where priced, dollars.

The default Wardrik.Authorizer. Allows what the scope's Wardrik.Statements allow.

Delivers trigger prompts to agents. Implements WardrikTriggers.Target.

Tools offered by workers: programs that connect to Wardrik and run tool calls in their own process, like MCP servers.

Raised when a worker goes away or times out during a call. The call may have run.

Raised when no worker offers a tool as it's called. The call didn't run.

Gives agents files: a tree of their own, shared trees, and files read into every turn.