Kimi Work by Moonshot AI

Medley vs. Kimi Work

Kimi Work targets knowledge workers with a powerful desktop agent and browser automation. Medley targets founders orchestrating Claude Code and Codex across coding and company-building missions.

Free · Mac OS

Kimi Work brings Moonshot’s Kimi K2.6 model — 1T parameters, 256K context — into a desktop agent aimed at finance, ops, and professional services. It handles coding plus docs and browser tasks with open API key access and Claw Groups for mid-swarm intervention.

Medley overlaps on “more than coding” but differs in architecture: model-agnostic routing across Claude Code and Codex, visible mission DAGs, a centralized attention queue, and a project layer designed for AI-native founders on Mac.

What Kimi Work does well

Kimi Work packages a frontier-class model into a local-first desktop experience for knowledge workers who need browser automation, document work, and long-horizon tasks — not just diffs.

BYOK is supported, and Claw Groups let you intervene mid-swarm when agents need redirection. For teams already standardized on Kimi’s ecosystem, the integrated agent experience is compelling.

Where Kimi Work stops

There is no true project abstraction — work is session-level and long-horizon, but missions are not decomposed into an editable DAG with dependencies and parallel routing across multiple agent types.

Model support centers on Kimi K2.6 rather than dynamic routing to Claude Code, Codex, and local models based on task type. The attention model uses Claw Groups, not a single cross-project queue of sign-off decisions.

What Medley does differently

Medley is agent-agnostic by design — Claude Code for depth, Codex for speed, others as you choose — with dynamic routing per sub-task. Missions are first-class objects with targets, budgets, and sign-off gates.

A centralized attention queue prioritizes what needs you across every project. Cost per finished task and decision memory compound your judgment over time.

Key differences

Model routing

Medley routes across Claude Code, Codex, and more. Kimi Work centers on Kimi K2.6.

Project decomposition

Medley produces editable mission DAGs. Kimi Work runs long sessions without project-level structure.

Attention queue

Medley unifies decisions across projects. Kimi uses Claw Groups within swarms.

Founder ICP

Medley is built for Claude Code / Codex power users shipping company-building work.

Full product comparison

MedleyKimi Work
Product typeProject orchestratorDesktop AI agent & orchestrator
Coding vs. non-coding workBoth — coding + GTM, content, ads, CROBoth — coding + finance, docs, browser
Project layerProject-level — task decomposition across the projectSession-level — long-horizon but no project abstraction
Attention queue / human-in-the-loopCentralized queue — humans pulled in only when neededClaw Groups for mid-swarm intervention
BYOK (bring your own key)Yes — fully BYOKYes — open API key access
Local vs. cloudLocal — desktop appLocal-first app, cloud model inference
Model supportClaude Code, Codex — dynamic routingKimi K2.6 (1T param, 256K context)
Platform / OSMac — desktop appmacOS (Apple Silicon), Windows
Pricing & business modelFree — BYOK, pay your own API costsFree tier; Moderato $19/mo; Vivace $199/mo
Target user / ICPFounders & power users of Claude Code / CodexKnowledge workers in finance, ops, professional services
Open sourceNoPartial — K2.6 model weights open; app closed

Data as of June 2026. Sources: web research, medley.sh.

Which should you choose?

Choose Kimi Work if

  • You want a desktop agent built around Kimi K2.6’s context and capabilities
  • Your work is finance, ops, or professional services with browser automation
  • You prefer Moonshot’s integrated stack over multi-agent BYOK routing
  • Claw Groups fit how you intervene in agent swarms

Choose Medley if

  • You already run Claude Code and Codex and want orchestration across both
  • You need mission DAGs, sign-off gates, and cost per task
  • You want one attention queue across coding and GTM missions
  • Vendor flexibility and dynamic model routing matter more than a single-model agent

Medley vs. Kimi Work FAQ

Both do non-coding work — what is the difference?
Both span coding and non-coding tasks. Medley adds project-level decomposition, multi-agent routing, and a founder-focused attention queue across Claude Code and Codex — not a single-model desktop agent.
Which platforms are supported?
Kimi Work targets macOS (Apple Silicon) and Windows. Medley is a Mac desktop app today, local-first with BYOK.
How does pricing compare?
Medley is free to download — you pay your own API costs. Kimi Work offers free tier plus Moderato ($19/mo) and Vivace ($199/mo) plans on top of model usage.

Orchestrate Claude Code and Codex on Mac.

Join the waitlist for Medley — mission control for founders who delegate real work to their agents.

Join the waitlist