kindling-client

Daemon-backed Rust SDK for kindling — local-first memory for AI-assisted development.

Use kindling-client when your Rust application, agent, editor extension or workflow tool needs to read and write kindling memory safely while other tools may be running at the same time.

This is the recommended Rust SDK for most integrations. It is a thin async client that speaks HTTP/1 over a Unix domain socket (TCP fallback on Windows), auto-spawns the daemon on first use, and re-exports the domain types so you depend on this crate alone.

Install

[dependencies]
kindling-client = "0.2"

Durable emit (`spool` feature)

The opt-in spool feature adds SpooledClient — a durable-emit layer that buffers observations to a local NDJSON spool when the daemon is unreachable and replays them on the next successful append or an explicit flush().

[dependencies]
kindling-client = { version = "0.2", features = ["spool"] }

There is no standalone kindling-spool crate. The spool module ships inside kindling-client behind features = ["spool"] (off by default). Enable it only when your producer must survive daemon outages.

Example

use kindling_client::{Client, CapsuleType, ScopeIds};

#[tokio::main]
async fn main() -> Result<(), kindling_client::ClientError> {
    let client = Client::new()?;

    let health = client.health().await?;
    println!("kindling daemon schema v{}", health.schema_version);

    let capsule = client
        .open_capsule(
            CapsuleType::Session,
            "investigate flaky test",
            ScopeIds::default(),
            None,
        )
        .await?;

    println!("opened capsule {}", capsule.id);

    Ok(())
}

When to use this crate

Use kindling-client when you want:

Daemon-backed access to kindling.
Safe concurrent integration from multiple tools.
A lightweight SDK without pulling in SQLite directly.
Domain types re-exported from one crate.

For embedded, single-process usage, use kindling-service instead.

How it fits into kindling

kindling-client talks to the daemon you start with kindling serve (from the kindling binary). The daemon owns the local SQLite store and keeps access safe when several tools touch the same project memory at once.

Documentation

Full docs: docs.eddacraft.ai/kindling

Relevant guides:

Licence

Apache-2.0

kindling-client 0.3.0