Rig

Rig is a Rust framework for LLM agents. rig-treeship wraps any Rig PortableTool so that every call it makes leaves a signed, chained, offline-verifiable receipt. Both stacks are Rust, so this is a type-level integration: signing happens in-process with Ed25519 over DSSE envelopes through treeship-core, and the compiler checks the seam.

The crate lives in the Treeship monorepo at packages/rig (it moved in with #292 in August 2026) and publishes to crates.io in lockstep with every Treeship release, so rig-treeship and treeship-core always share a version.

cargo add rig-treeship

What it does

use std::sync::Arc;
use rig_treeship::{AttestedExt, TreeshipLedger};

let ledger = Arc::new(TreeshipLedger::open_default("agent://my-agent")?);
let tool = MyTool.attested(ledger.clone());
// register `tool` with your runtime exactly like the bare tool

Every call then:

1. hashes the deserialized arguments (SHA-256, compact JSON),
2. runs the inner tool,
3. hashes the output, or records the failure as outcome: "error",
4. signs a treeship/action/v1 statement carrying both hashes, chained to the previous receipt via parentId,
5. returns the result only after the receipt is stored. An unattested action is treated as no action.

The wrapper is transparent to the model: same name, description and JSON schema. It changes what the agent can prove, not what it can do.

Verifying

let head = ledger.head().unwrap();
let verified = ledger.verify_chain(&head)?;   // walks parentId back to genesis
println!("{} receipts verified", verified.length);

verify_chain re-derives every artifact id from its PAE bytes, checks each Ed25519 signature, and confirms the stored parent pointer matches the signed parentId. Receipts are standard Treeship artifacts on disk, so the CLI applies too: treeship verify <id>, and treeship hub push <id> for a public URL.

Where to go next

• packages/rig: the crate, with examples/attested_agent.rs and an interop test against the CLI verifier
• rig-treeship on crates.io
• treeship verify