PRE-ALPHA DEVELOPMENT
KRNL HARNESS

The trusted layer around autonomous agents

Built with Axum, Tokio, and SQLx in Rust: persistent sessions, event-driven scheduling, capability-scoped tools, human approvals, isolated workers, and a record of every privileged move.

Isometric Karnelian harness workshop
CURRENT CAPABILITIES

Built for controlled autonomy

The Harness provides the infrastructure an agent needs to work continuously without becoming an opaque black box.

Persistent sessions
Semantic memory
Capability-bounded Skills
Human approval gates
Tamper-resistant Ledger
Event-driven scheduling
Local model routing
Isolated worker runtimes
RUNTIME MODEL

Intent enters, governed work leaves

Lian maintains the relationship with you while the orchestrator turns intent into persistent tasks, event updates, approval requests, and bounded worker execution.

01

Interpret

A soul and persistent session give the request identity, goals, and continuity.

02

Plan

Context assembly combines memory, live state, and the relevant active skills.

03

Authorize

Policy evaluates required capabilities and pauses sensitive actions for approval.

04

Execute

A specialized worker performs the task and streams meaningful events back.

Lian dispatching work to isolated specialist runtimes
EVENT CORE

Continuous, not stateless

The orchestrator exposes REST endpoints and a WebSocket event stream while the scheduler keeps long-running work and recurring behavior alive.

  • Persistent sessions survive individual prompts
  • HeartbeatTick events are sampled to keep telemetry useful
  • Task state, approvals, and execution progress stay observable
  • Experience progression uses a level-scaled XP curve
SKILL BOOK

Capabilities with contracts

More than fifty active skills declare their runtime and authority needs before execution. A much larger migration queue is tracked separately and is not presented as shipped.

  • Node.js, Python, WebAssembly, and native Rust workers
  • Capabilities and resources declared in manifests
  • Quantum research is an explicit skill category
  • Refreshable active Skill Book for discoverable tools