API Reference wardrik_protocol v#0.1.0-rc.1

What an agent is doing now.

An agent, and the operations on one.

Parses agent IDs.

Operations on every agent the scope can see.

An amount in US dollars (usd), tokens (tokens), or both. A limit leaves out what it doesn't limit.

Operations on the deployment's audit log.

Where a person signs in for a call that waits on it: the url, and the code for a device sign-in.

A tool call an agent is running: its id, the tool's name, and elapsed_ms.

Operations on an org's ceilings: the budgets its agents share.

A behaviour for chat apps, like Slack, that people reach agents through.

A channel added to one workspace, such as a Slack app added to one Slack workspace.

A behaviour for storing the workspaces a channel is added to, each with its token.

A channel's response to its chat app: the status, headers and body to send unchanged.

The result of connecting an account. If it isn't connected yet, authorization says where the owner signs in.

An account an owner can connect: its provider, and whether it's connected.

Operations on the accounts an org or person connects for their agents, by provider.

A decided call: its id, and its decision, :approved or :rejected.

What an agent is and what it may do: an agent file's frontmatter, as a struct.

Operations on the deployment: listing its nodes and reloading every agent's definition.

An operation's failure, as the protocol's error.

One event, as a CloudEvent.

Operations on the events of every agent the scope can see.

The result of forgetting many agents. forgotten lists the IDs forgotten. kept and failed list {id, why} pairs.

Delivers a service's request to an agent's webhook trigger.

A call an agent waits on a person for.

Operations on the calls waiting on a person, across every agent the scope can see.

An MCP server an owner added for their agents: its name, url, and auth. auth is how agents sign in to it: :oauth, :secret or :none.

The result of adding an MCP server. If it isn't connected yet, authorization says where someone signs in first.

Operations on the MCP servers an org or person adds for their agents at runtime.

A message sent to an agent: its id, and the cursor its answer comes after.

One of the deployment's nodes: its name, and how many agents run on it.

The deployment's nodes, and how many agents are asleep.

Tells someone what an agent did while nobody watched.

Converts each operation's input and output between Elixir and the protocol's JSON.

A page of events, and next, where to read on from.

An agent's answer: its text, the message it answers, and the run that wrote it.

Names what an action acts on, as IAM resources do.

One run of an agent: how it ended, how long it took, and what it used and cost.

An agent's runs, newest first, and their totals.

Totals across an agent's runs.

Who is acting through Wardrik, and what they may do.

A secret an owner holds, by name. kind is :secret for one they sealed, or :grant for an account they connected.

Operations on an owner's secrets: sealing one, listing them by name, and removing one.

A behaviour for what runs Wardrik's operations.

What an org's agents have spent against one ceiling in its current period. See Wardrik.Ceiling.

Allows actions on the resources its patterns match, like an IAM statement.

An agent's conversation as a person reads it, oldest first.

An API token, without its secret.

A new token and its secret. The secret is returned only once. Send it as Authorization: Bearer <secret>.

Operations on the deployment's API tokens. People and programs use them to reach Wardrik over HTTP.

Asks an agent something without anyone typing. A trigger fires

Where a service sends a trigger requests. secret is the secret it holds, as ${NAME}. verify is how a request proves it: :shared or :standard.

Wardrik's wire format: events as CloudEvents, and operations as OpenAPI.