Sakana Fugu by Sakana AI
Medley vs. Sakana Fugu
Sakana Fugu routes complex work through a managed cloud API with strong benchmarks. Medley shows you the plan, runs locally in Claude Code, and attributes cost to every finished task.
Paste into Claude Code
Sakana Fugu is a managed multi-agent cloud service — a small TRINITY coordinator routing across a frontier model pool. It targets enterprise users and researchers tackling multi-step problems with an OpenAI-compatible API.
Medley represents the opposite architectural bet: local execution, visible DAG decomposition, the models you already have, per-task cost receipts, and a human attention queue — no black-box routing, no managed billing layer.
What Sakana Fugu does well
Fugu delivers strong software engineering benchmark results and handles code review plus research-style tasks through a drop-in API. For teams that want managed infrastructure and minimal integration work, the serverless model is attractive.
Pricing tiers (Standard $20/mo, Max $200/mo) package access without operating your own orchestration layer. Multi-agent parallelism happens in the cloud without local setup.
Where Sakana Fugu stops
Routing is opaque — you cannot see which model handled which sub-task, edit the execution plan, or interrupt mid-run with full audit trail. There is managed keys only; you pay Sakana’s managed rates, not your negotiated provider pricing.
Fully autonomous by design — no centralized attention queue for human sign-offs. Regional access has been limited (EU/EEA blocked as of research date). Data passes through Sakana’s cloud, not your Mac.
What Medley does differently
Medley runs on your Mac with the models you already have. It is designed to make every sub-task visible in an editable DAG before and during execution, so humans enter through an attention queue rather than being excluded from the loop.
Cost per completed task, decision memory, and missions that span coding and adjacent company-building work aim to address the gaps left by a black-box API optimized for benchmark scores.
Key differences
Transparency
Medley shows the DAG and routing decisions. Fugu routes invisibly through managed infrastructure.
Claude Code plugin
Medley is local-first with the models you already have. Fugu is cloud-only with managed keys only.
Human-in-the-loop
Medley queues decisions that need you. Fugu runs fully autonomous.
Cost model
Medley: per-task receipts to your providers. Fugu: per-token through Sakana.
Full product comparison
| Medley | Sakana Fugu | |
|---|---|---|
| Product type | Project orchestrator | Managed multi-agent cloud API |
| Coding vs. non-coding work | Both — coding and adjacent company-building work | Both — code review + research tasks |
| Project layer | Project-level — task decomposition across the project | Session-level — black-box API, no DAG |
| Attention queue / human-in-the-loop | Centralized queue — humans pulled in only when needed | Fully autonomous — no human intervention |
| Model access | Uses models you already have in Claude Code | No — managed service, managed keys only |
| Local vs. cloud | Local — Claude Code plugin | Cloud-only — serverless API |
| Model support | Models you bring into Claude Code (Claude, GPT, Gemini, and more) | 0.6B TRINITY routing a frontier model pool |
| Platform / OS | Mac — desktop app | Platform-agnostic API (web console + OpenRouter) |
| Pricing & business model | Free — models you already have in Claude Code | Standard $20/mo; Max $200/mo |
| Target user / ICP | Founders & power users of Claude Code | Enterprise users, researchers, multi-step problem solving |
Data as of June 2026. Sources: web research, medley.sh.
Which should you choose?
Choose Sakana Fugu if
- You want a managed OpenAI-compatible API with no local setup
- Opaque routing is acceptable for your compliance and trust model
- You are outside restricted regions and can access the service
- Benchmark-optimized coding agents are your primary workload
Choose Medley if
- You want multi-agent missions inside Claude Code
- You want visible decomposition and per-task cost attribution
- Human sign-off gates and an attention queue are required
- You orchestrate Claude Code across code and business work
Medley vs. Sakana Fugu FAQ
- Is Fugu easier to integrate?
- Fugu’s OpenAI-compatible API is drop-in simple. Medley is a desktop orchestration product — different integration model, different control tradeoff.
- Is Fugu available in the EU?
- As of June 2026 research, Fugu was blocked in the EU/EEA. Medley is a local-first Mac app — install at medley.sh/install.
- Does Fugu have better coding benchmarks?
- Fugu reports very strong SWE-bench results. Medley optimizes for founder workflows — cross-domain missions, transparency, and cost per task — not benchmark leaderboard position.
Transparent orchestration on your Mac.
Install Medley — no black box, no managed markup, the models you already have.
Install MedleyI