Struct Client 

pub struct Client { /* private fields */ }

Async gateway client. Share it behind an Arc; methods take &self.

Every call is POST {gateway_url}/api/<method> with the session’s bearer token and headers, after re-checking the URL against the Policy. A client created with a Cursor access token (Client::connect_cursor / Client::with_cursor_token) re-runs EnsureSandBox once and retries when the gateway answers 401 or 417; persist the new session from Client::take_refreshed_session.

use grokbot_api::{Client, ClientOptions, Session};

let client = Client::new(
    Session::new("https://box.us1.cursorvm.com", "gateway-token"),
    ClientOptions::default(),
)?;
for agent in client.list_agents().await? {
    println!("{} {}", agent.id, agent.display_name());
}

Implementations

impl Client

pub fn new(session: Session, opts: ClientOptions) -> Result<Self>

Client for an existing gateway session. The URL is checked against the policy on every call.

pub fn with_cursor_token( session: Session, cursor_token: impl Into<String>, opts: ClientOptions, ) -> Result<Self>

Like Client::new, but keeps the Cursor access token so a 401 triggers one EnsureSandBox re-exchange and a retry.

pub async fn connect_cursor( access_token: &str, opts: ClientOptions, ) -> Result<Self>

EnsureSandBox then build a client that can re-exchange.

pub fn session(&self) -> Session

The current session (it changes after a 401/417 re-exchange).

pub fn take_refreshed_session(&self) -> Option<Session>

The new session if a 401 re-exchange happened since the last call (persist it).

pub async fn call_raw(&self, method: &str, body: Value) -> Result<Value>

Call any gateway method with a raw JSON body and return the raw JSON response. Handles the 401/417 re-exchange like the typed methods.

pub async fn list_agents(&self) -> Result<Vec<Agent>>

listAgents {} → bots and groups (hidden ones included; check hidden_from_sidebar).

pub async fn create_agent(&self, input: &CreateAgent) -> Result<Option<Agent>>

createAgent {name, description, title, avatarShape, avatarColor, origin: "user"}.

pub async fn update_agent( &self, id: &str, profile: &AgentProfile, ) -> Result<Value>

updateAgent {id, profile: {name, description, title?, avatarShape?, avatarColor?}}.

pub async fn delete_agent(&self, id: &str) -> Result<Value>

deleteAgent {id}.

pub async fn set_agent_notify_on_updates( &self, id: &str, enabled: bool, ) -> Result<Value>

setAgentNotifyOnUpdates {id, isEnabled}.

pub async fn set_agent_hidden_from_sidebar( &self, id: &str, hidden: bool, ) -> Result<Value>

setAgentHiddenFromSidebar {id, isHidden}.

pub async fn create_group(&self, input: &CreateGroup) -> Result<Option<Agent>>

createGroup {name, description, memberAgentIds} (1..=6 distinct members, no nested groups).

pub async fn set_group_members( &self, group_id: &str, member_agent_ids: &[String], ) -> Result<Option<Agent>>

setGroupMembers {id, memberAgentIds}.

pub fn set_agent_session(&self, agent_id: &str, session_id: &str)

Route calls for agent_id through session_id (a shared agent’s viewer session), sent as sessionId on sendPrompt, promptAcceptanceStatus and getAgentTranscriptTail. Client::list_team_agents sets this automatically.

pub async fn list_team_agents(&self) -> Result<Option<Vec<TeamAgent>>>

Cursor backend GrokBotService/ListGrokBotTeamAgents: your agents plus team agents shared with you. Ok(None) when this client has no Cursor access token (a plain gateway session). Shared agents’ viewer sessions are remembered for later calls.

pub async fn send_prompt(&self, req: &SendPrompt) -> Result<SendReceipt>

sendPrompt {agentId, prompt, clientNonce, replyToId?, sessionId?} → receipt only.

Replies arrive later: poll Client::transcript_tail. On error use Error::delivery: Unknown means the prompt may have landed, so check the transcript or Client::prompt_acceptance_status before resending. A gateway refusal ({status: "refused"} / {accepted: false}) is returned as Error::Http with status 422.

pub async fn prompt_acceptance_status( &self, agent_id: &str, client_nonce: &str, ) -> Result<AcceptanceStatus>

promptAcceptanceStatus {agentId, clientNonce}: the gateway’s own lookup of whether a prompt with this nonce landed (what the desktop app uses when a send’s outcome is uncertain).

pub async fn transcript_tail( &self, id: &str, limit: u32, before_seq: Option<u64>, ) -> Result<Transcript>

getAgentTranscriptTail {id, limit, beforeSeq?} (limit clamped to 1..=200).

Omitting before_seq returns the newest page. When the response includes nextBeforeSeq, pass it here to walk older pages (as the desktop app does).

pub async fn transcript_history( &self, id: &str, page_size: u32, max_entries: usize, ) -> Result<Transcript>

Walk getAgentTranscriptTail pages until the chat is exhausted or max_entries is reached. Newest page first; returned entries are oldest → newest.

pub async fn thread(&self, id: &str, root_id: &str) -> Result<Transcript>

getAgentThread {id, rootId}.

pub async fn get_agent_workflows(&self, id: &str) -> Result<Vec<Workflow>>

getAgentWorkflows {id}. The library is shared per box; id is any bot id.

pub async fn import_agent_workflow_text( &self, id: &str, markdown: &str, ) -> Result<ImportResult>

importAgentWorkflowText {id, markdown}.

pub async fn delete_agent_workflow( &self, id: &str, workflow_id: &str, ) -> Result<Value>

deleteAgentWorkflow {id, workflowId}.

pub async fn resolve_local_tool_permission( &self, agent_id: &str, entry_id: &str, request_id: &str, resolution: LocalToolResolution, ) -> Result<Value>

resolveLocalToolPermission {agentId, entryId, requestId, resolution: "allow-once"|"deny"}. Only call on an explicit user decision.

pub async fn resolve_auto_review_approval( &self, agent_id: &str, entry_id: &str, request_id: &str, resolution: AutoReviewResolution, ) -> Result<Value>

resolveAutoReviewApproval {agentId, entryId, requestId, resolution: "approved"|"denied"}. Only call on an explicit user decision.

Trait Implementations

impl Debug for Client

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.