Why they built it
Marsh saw static analysis consuming time in a large Python codebase and believed rebuilding the tooling in Rust could remove that delay. He built Ruff to test that theory that a faster checker that preserved existing standards and conventions would be adopted immediately. Early adoption of Ruff and its formatter gave him reason to broaden into neighboring friction with package management with uv and type checking with ty.
How it ended
Astral did not fail from lack of use; OpenAI announced an agreement on March 19, 2026 to acquire Astral with the team intended to join Codex toward systems that can participate in the entire development workflow. Its free tools had hundreds of millions of monthly downloads, but its paid registry pyx was still in beta and disclosed no revenue in the sources reviewed. The evidence supports a strategic acquisition before its independent commercial model was proven, not a shutdown.
What went wrong
Most important first.
- 1
Paid product never showed revenue Monetization
Astral introduced pyx in beta in August 2025 with Ramp, Intercom and fal as early partners but disclosed no contract values, paid conversion or revenue, and its page as of October 2026 no longer accepted new signups.
- 2
Huge free use did not equal a business Product-market fit
Astral reported more than 100 million monthly installs and more than 500 million daily requests powered by uv in August 2025, which are described as distribution and workload measures, not paying accounts or revenue.
- 3
Strategic value arrived before commercial proof Other
OpenAI described moving toward systems that can participate in the entire development workflow, making Astral's environment for running and checking Python code valuable to an AI platform before standalone profitability was established.
- 4
Toolchain maintenance required ongoing work Technology
Astral's August 2025 ZIP-parsing advisory and later protected release environments, artifact attestations and controls showed mature tooling required sustained incident response and release work beyond the first implementation.
What could be different today
- OpenAI has stated Codex should run tools and verify software throughout development, so automated coding jobs need evidence of which dependencies and checks actually ran.
- Astral shipped an experimental October 2026 integration where uv and ty work together to check scripts against declared requirements and run project checks through uv check.
- Astral now documents artifact attestations, protected release environments and controls on privileged automation, raising the baseline for execution evidence a rebuild must complement.
- pyx no longer accepts new signups as of October 2026, so teams cannot newly adopt Astral's hosted registry path.
