Windsurf
CodeiumAI-native IDE, forked from VS Code, built around an agentic flow engine that keeps context across multi-step coding tasks and suggests and executes terminal commands, file edits and debugging steps. It runs multi-model and multi-agent over the Agent Client Protocol, blending inline suggestions with autonomous agent actions.
Formerly Codeium, then Windsurf; acquired by Cognition in 2025 and now shipping as Devin Desktop, with windsurf.com redirecting there.
Openness
1 high confidence- source
- closed
- license
- proprietary(SaaS, Cognition)
- self-host
- no
Proprietary agentic IDE; now part of Cognition's Devin Desktop. Closed source, subscription SaaS.
- https://cognition.ai/blog/windsurf recorded 2026-08-13
Cognition's own acquisition announcement - "the acquisition includes Windsurf's IP, product, trademark and brand"
- https://devin.ai/desktop recorded 2026-08-13
Windsurf shipped as Devin Desktop, a proprietary agentic IDE sold on Free/Pro/Teams/Enterprise tiers; no source and no self-host option
Adoption
4 medium confidenceCognition's product page states a user total and an enterprise-customer count without a measurement period, so neither is an active-use figure. They show the scale the vendor reports, not measured usage.
- https://cognition.ai/blog/windsurf recorded 2026-09-29
Acquisition announcement lists "$82M of ARR and a fast-growing business, with enterprise ARR doubling quarter-over-quarter".
- https://devin.ai/desktop recorded 2026-09-29
Stats block shows "1M+" beside "Trusted by over a million developers worldwide" and "4000+" enterprise customers, with no measurement period attached.
Capability
4 medium confidenceWindsurf has a mature agentic-IDE feature set: multi-agent orchestration and model-agnostic operation through the Agent Client Protocol. It publishes no first-party SWE-bench score, so it is judged on this feature set, placing it a notch below OpenHands, which does have a verified benchmark result.
- https://devin.ai/desktop recorded 2026-09-29
Product page describes "Work across models and agents, powered by the Agent Client Protocol (ACP)." and "Fast Context finds the exact files and lines your agent needs" plus "Supercomplete predicts your next thought, not just your next edit."; searched for SWE-bench: none.
Verified 2026-09-29, except some scores, still awaiting re-confirmation