AGPL-3.0 · Local-First · Provider-Agnostic

Distributed Workspace
Orchestration

The AI-powered workspace that runs on YOUR machine,
not someone else's cloud.

Request Beta Access
shikki
$ shikki spawn --agents 3 --mesh local
▸ agent-0 connected via NATS (latency: 0.4ms)
▸ agent-1 connected via NATS (latency: 0.3ms)
▸ agent-2 connected via NATS (latency: 0.5ms)
✓ mesh ready — 3 agents, 0 pending tasks
$ shikki dispatch "refactor auth module"
▸ dispatching to agent-0 (planner)
▸ dispatching to agent-1 (coder)
▸ dispatching to agent-2 (reviewer)
✓ task completed in 47s — 3 files changed

Built for developers who own their stack

Multi-Agent Dispatch

Spawn, coordinate, and monitor multiple AI agents working in parallel. Planner, coder, reviewer — each with dedicated context.

NATS Mesh

Sub-millisecond agent communication over NATS. Local or distributed — your agents talk directly, no cloud relay.

TUI Dashboard

Real-time terminal interface. Watch agents work, inspect task queues, and intervene when needed — all from your terminal.

Native iOS & macOS

Swift-native client for monitoring and dispatching from anywhere. Push notifications, Shortcuts integration, offline-capable.

Provider-Agnostic

Anthropic, OpenAI, Ollama, local models — plug in any LLM provider. Switch without changing your workflow.

Local-First

Your code never leaves your machine unless you say so. Full sovereignty over data, context, and model routing.

How it works

01

Describe

Give Shikki a task in natural language. It breaks it down into a plan with roles for each agent.

02

Dispatch

Agents receive their assignments over the NATS mesh. Each operates in an isolated workspace with scoped context.

03

Deliver

Results are merged, reviewed, and presented. You approve, iterate, or ship — always in control.

Get early access

Shikki is in closed beta. Drop your email and we'll reach out when a spot opens.