AI Deck

Autohand — A self-evolving AI coding agent platform with IDE extensions, a CLI, and in-house models

An AI coding agent platform from Autohand AI Limited in New Zealand. It bundles IDE extensions for VSCode and Zed, a CLI that runs in the terminal, Commander for directing several agents at once, and Evolve, a self-improving agent that refines its own workflow as you use it. What sets it apart is that it does not lean entirely on outside LLMs: it ships its own models — Moa, the flagship model for deep reasoning, and Fantail, a small and fast model. Since its release in February 2026 it has kept expanding into a foundation for running agents, including local execution and MCP integration.

Key Features

  • Multiple entry points: IDE extensions, CLI, and SDK: Extensions for VSCode and Zed, plus a CLI that stays in the terminal (autohandai/code-cli). Browser, Slack, and SDK workflows are supported as well, so the same agent is available inside and outside the editor
  • Evolve, a self-improving agent: Rather than treating every run as a one-off, it updates its procedures and how it handles context based on the results of past work. The more you use it on one repository, the closer it moves to that project’s conventions
  • Commander for multi-agent management: A management layer for running several agents in parallel and reviewing their progress and output in one place. It pays off on larger changes you want to split into parallel tasks
  • Moa and Fantail for different jobs: Moa is the flagship model, built around showing the evidence behind high-risk changes; Fantail is a small, fast model suited to local-first work. Open weights for Fantail have been announced under CC BY 4.0
  • Cloud and local execution: Moa and Fantail can be used through the cloud, and local execution via MLX on Apple Silicon is also supported, which leaves an option for environments where code must not leave the machine
  • MCP support: External tools and data sources can be connected through the Model Context Protocol, extending an agent’s reach into the environment around the project

Pricing

PlanMonthly priceWhat’s included
Free$0Connect your own LLM provider (BYOK)
Hosted (lower tier)$8Request allowance for Autohand-hosted models
Hosted (upper tier)$39A larger request allowance
EnterpriseContact salesUnlimited agents, private models, SOC 2 compliance

Pricing is as of August 2026. Plan names and included allowances may change, so please check the official pricing page for the latest details.

Pros & Cons

Pros

  • The free plan lets you bring your own API key, so anyone who already pays for an LLM can try it at no extra cost
  • With IDE extensions, a CLI, and an SDK, you can choose where it fits into your existing development flow
  • Because it has its own models, you can pick either an all-in-one setup or a BYOK setup
  • Local model execution on Apple Silicon is supported, which makes a no-code-leaves-the-machine setup worth considering
  • Paid plans start at $8 per month, which is easy for an individual developer to justify

⚠️ Cons

  • Released in February 2026, it is still new, with a limited track record and little Japanese-language material
  • There is not much objective evaluation comparing Moa and Fantail against the established large models
  • Plan names and request allowances need to be confirmed on the official site; the public information alone makes the limits hard to pin down
  • The breadth of Evolve and Commander can be overkill for someone who only wants code completion

Comparison with Similar Services

ItemAutohandClaude CodeCursorCline
Form factorIDE extensions + CLI + SDKCLI-centeredAI-integrated editorVSCode extension
ModelsIn-house (Moa / Fantail) + BYOKAnthropic modelsChoice of several modelsBYOK (connect vendor APIs)
Local executionApple Silicon (MLX) supportedCloud-basedCloud-basedLocal models can be connected
Multi-agentManaged through CommanderSubagent featureLimitedLimited
Free usageFree with BYOKSeparate from free chat planFree tier availableExtension is free (API costs separate)

Who Is It For

  • Developers who already hold LLM API keys and want to swap only the agent tooling
  • People who work across both VSCode and Zed, or move back and forth between terminal and editor
  • Teams who want to split a larger change into tasks and run several agents on them in parallel
  • Anyone who cannot send code outside their machine and wants to try coding assistance with local models
  • People who like touching new tools early to judge whether they fit their workflow

Summary

Autohand pulls IDE extensions, a CLI, multi-agent management, a self-improving agent, and in-house models into one broad AI coding platform. Starting free with BYOK and moving to hosted models when you need them is an easy path for an individual developer. It is still early days, though, and the real strength of its own models and the practicalities of long-term use are only beginning to show, so the realistic way in is to try the free plan on your own repository and compare it against the coding agent you use today.

← Blog