Fractal Works, Inc.
Fractal Works makes AutoRouter, a Mac-based model router for coding agents. AutoRouter chooses a model for each coding task, prioritizes users' existing Claude, ChatGPT, Grok and Gemini subscriptions, and uses lower-cost or additional models when appropriate.
At a Glance
- Individual developers using AI coding agents
- Engineering teams using Codex and Claude Code
- Organizations managing AI coding spend and model policy
- Teams with multiple Claude, ChatGPT, Grok or Gemini subscriptions
- +1 more
AI Tools by Fractal Works, Inc.
(1)AutoRouter
Model Router for macOS Coding
Discussions
No discussions yet
Be the first to start a discussion about Fractal Works, Inc.
Latest News
AutoRouter published guidance for using multiple Claude accounts in Claude Code, including automatic account failover when one account reaches its limit.
AutoRouter published its OpenCode vs Claude Code comparison, positioning AutoRouter as a way to bring OpenCode-style model choice into Claude Code.
AutoRouter documented using Claude in Codex and GPT in Claude Code through connected subscriptions without keys or configuration.
AutoRouter published support guidance for MiniMax M3.1 Flash in Claude Code and Codex.
Products & Services
A signed and notarized macOS 14+ app that connects Codex, Claude Code, Cursor, Claude Desktop, T3 and Terminal to a model-routing service. It adds 20+ models to model pickers, routes each turn, and uses connected subscriptions before paid credits.
A ChatGPT app that recommends models for coding tasks and, after account connection, shows routing activity and lets users edit routing rules; it does not run prompts or read ChatGPT conversations.
A team deployment and governance offering with Jamf/MDM rollout, organization rules, model controls, budgets, spend reporting by team/project/tag, zero data retention, SSO, SCIM and audit logs.
Market Position
AutoRouter positions itself as a plan-aware model router embedded in existing coding tools rather than a generic API gateway. Its differentiators are using subscriptions customers already pay for, automatically switching among accounts, adding model choice inside Codex/Claude Code/Cursor, and combining per-turn routing with enterprise budgets and retention controls. Its publicly discussed alternatives and adjacent competitors include OpenRouter, LiteLLM, Vercel AI Gateway, Claude Code Router, OpenCode, Cursor, GitHub Copilot CLI, Aider and Cline.
Leadership
Founders
Tyler J. Cagle
Cofounder of AutoRouter; previously worked at Capital (acquired by Rho) and cofounded Crypto: The Game (acquired by Uniswap Labs in 2024).
Nadav Hollander
Cofounder and CEO of AutoRouter; previously CTO of OpenSea and CEO of Dharma.
Executive Team
Nadav Hollander
Cofounder and CEO
Previously CTO of OpenSea and CEO of Dharma.
Tyler J. Cagle
Cofounder
Previously worked at Capital and cofounded Crypto: The Game; his prior ventures were acquired by Rho and Uniswap Labs respectively.
Founding Story
AutoRouter was started to address the waste and inconsistent results caused by sending every coding-agent prompt to a frontier model. The initial vision was an optimization layer that understands a request, selects the best-fit model for that task, and saves money while working inside the coding tools developers already use.
Business Model
Revenue Model
Free routing across up to two connected subscriptions; paid Pro subscription for unlimited subscriptions and models outside those plans, with usage charged from included/top-up API credits at provider list price plus 10%; Enterprise is sold as a custom team/seat offering.
Pricing Tiers
Up to two Claude, ChatGPT, Grok or Gemini subscriptions, personal routing rules and supported harnesses; models outside connected plans are excluded.
Unlimited subscriptions, every model via API credits, $20 in included monthly API credits, priority support and early access. New accounts receive a seven-day Pro trial or until $20 of usage is spent.
Team/seat pricing with Jamf or MDM deployment, organization rules/models/budgets, spend reporting, zero data retention, SSO, SCIM and audit log.
Target Markets
- Individual developers using AI coding agents
- Engineering teams using Codex and Claude Code
- Organizations managing AI coding spend and model policy
- Teams with multiple Claude, ChatGPT, Grok or Gemini subscriptions
- Software engineering and AI infrastructure/MLOps users
- Coding-agent sessions in Codex and Claude Code
- Routing routine edits, file reads and tests to cheaper capable models
- Using Claude plans inside Codex and ChatGPT plans inside Claude Code
- Extending the usable capacity of multiple paid AI subscriptions
- Providing one model picker across several coding tools
- Enterprise governance of coding-model access, budgets, deployment and data retention