ctxmesh 0.1.0-beta.8

Rust SDK for ctxmesh, the Kubernetes-native control plane for AI agents. Typed clients for memory, knowledge, skills, feedback, delegation and agent-to-agent calls. Reads its endpoints from the environment the platform injects, so your code never holds credentials. Source, docs and issues: https://github.com/ctxmesh/ctxmesh
Documentation

ctxmesh — Rust SDK

Typed clients for agents running on ctxmesh, the Kubernetes-native control plane for AI agents.

Your agent runs in a pod beside the platform's sidecars. This crate is the typed way to reach them — conversation memory, long-term memory, knowledge bases, skills, feedback, agent-to-agent calls, delegation and handoff.

Your code never holds credentials. Endpoints and identity arrive in the environment the platform injects, so there is no API key to manage and no base URL to configure.

[dependencies]
ctxmesh = "0.1.0-beta.8"

Use it

use ctxmesh::{Client, Entry};

let cx = Client::from_env()?;               // reads MEMORY_PORT, CONVERSATION_ID, …

cx.memory_append(&Entry { role: "user".into(), content: "what changed?".into() }, None)?;
let history = cx.memory_get(None)?;

let facts = cx.search_agent("contact preference", 5, 0.0)?;
let hits = cx.knowledge_search("rollback procedure", None, 5)?;
cx.feedback(&trace_id, "helpfulness", 1.0, Some("clear"))?;

Client::from_env() returns Error::NotInPod outside the platform. Config::for_test(..) builds one explicitly.

Errors say which kind of "no" it was

  • Error::NotWired — the platform did not grant this capability. The port is absent because nothing is listening; a configuration answer, not a failure.
  • Error::Denied — a 403. The plane understood the call and refused it (a guardrail, a budget, the delegate fence), and it carries the reason.
  • Error::Api / Error::Transport / Error::Decode — everything else, kept apart so a caller can tell a refusal from an outage.

Blocking, and no async runtime

Built on ureq, not reqwest. This is an SDK: it is installed beside an application that has its own runtime and its own opinion about which one. A client that drags in tokio makes that choice for them.

Conformance tier

plane-client — every launcher route is reachable here. The managed agent loop and model client are authoring-tier and live in the Python and TypeScript SDKs. Every capability is also a plain HTTP endpoint, so this crate is convenience, never a requirement.

Documentation

Apache-2.0. Contributor and toolchain notes live in the repository.