Ed

Embed Ed. Your app gets a brain. Your users keep their data.

Ed is an agent you drop into an application. It loads an open-weights model, keeps the conversation, calls only the host tools you register, and hands back replies. The model runs on the user's device — there is no metered inference API and private context does not need to leave the device.

Why it exists

Seven applications needed the same thing: load a model, know whether it's ready, keep a conversation, hand back a reply. Seven times somebody wrote it again. Two of them ended up byte-identical in three files.

None of that code was interesting. It was the wiring between an inference engine that reports status by polling and a user interface that needs to be told — plus the flattening of Result<InferenceResult, _> into one payload a view can render either way. Ed is that layer, written once.

It deliberately doesn't own sessions, persistence, model catalogues, or downloads. Those differ per application and stay yours.

Onde runs the model. Ed makes it a feature.

Onde is the inference engine: weights, quantisation, Metal and CPU backends, tokens out. Ed is the part above it your users touch.

See it live

• smbCloud MailX — siMail drafts replies, summarises long threads, and helps triage the inbox. The mail never leaves the machine to get read.
• Siti AI — a private assistant for macOS, iOS, and Android, and the open reference app: its source is public, so it's the codebase to read for a full integration.
• Pepak Basa Jawa Komplit — Joko, a Javanese tutor who answers offline, which matters where the app is used.
• Rumi Learn Persian — Rumi, a Persian tutor, same model, same device.

Quick start

[dependencies]
ed-agent = "1.2"

use ed_agent::{Ed, GgufModelConfig};

let ed = Ed::new();
ed.load(GgufModelConfig::qwen25_1_5b(), None, None).await?;

let reply = ed.send("Summarise this thread.").await;
println!("{}", reply.reply.unwrap_or_default());

Stream a reply with Ed::stream (plain chat) or let Ed::run forward deltas to EventSink::text_delta. Ed::restore_history loads a saved conversation, Ed::generate is a one-shot call that leaves the history alone, and Ed::set_system_prompt / set_sampling change the model's settings. These three need a loaded model. Persistence stays yours; implement SessionStore over whatever you already have.

Construct with Ed::with_agent to provide the executor for registered tools and the approval handler for mutating tools. Read-only tools run directly; mutating tools must receive AllowOnce or AllowForSession first.

Crates

Crate	What it's for
ed-agent	The agent. No framework dependency.
ed-agent-ffi	UniFFI bridge used by the Swift XCFramework.
ed-agent-tauri	Tauri bindings: a sink that emits to the webview, the commands it invokes, and EdSessions for apps with a chat history and webview-answered approvals.

The Onde Agent Platform crates, for agents that editors drive over the Agent Client Protocol (ACP). SplitFire Agent is built on them.

Crate	What it's for
ed-acp	An ACP v1 server: sessions, auth, the turn loop on Onde Cloud, tool approval, workspace tools and MCP. A product supplies a Profile.
ed-acp-tui	A terminal UI for any ACP agent.
ed-mcp	MCP client glue over rmcp, the official Rust SDK: stdio and streamable HTTP servers, bearer tokens, tool namespacing, timeouts, progress and cancellation. With the agent feature, McpExecutor serves a toolset to Ed as tools.
ed-onde-account	Onde Inference account client: sign in, activate an app and get its Onde Cloud API key through ondeinference.com, with no Onde secrets in the host app.
ed-acp-testkit	Mock Onde Cloud endpoint, raw ACP client, the ACP registry probe and the ACP v1 conformance suite.

Free the intelligence.

Copyright

© 2026 Onde Inference. Licensed under MIT or Apache-2.0.