Recursor

Recursor is the earlier company/product direction behind [[FeatherlessAI|Featherless AI]] in Featherless AI: When Your Weekend Experiment Makes More Than Your Startup. The source says Recursor worked around the RWKV open-source project and let users fine-tune RWKV models with their own data.

Recursor matters in the wiki as the pre-pivot case. The team built GPU Hot Swapping because customers were creating many fine-tuned models and the company could not afford dedicated GPU capacity for every model. That internal resource constraint later became reusable infrastructure for serving Llama and [[MistralAI|Mistral]] models.

Connections

• Featherless AI - company direction that followed the pivot.
• Eugene Chia - founder explaining the transition.
• RWKV - original open-source model project around which Recursor was built.
• GPU Hot Swapping and AI Inference Cost Structure - internal infrastructure problem that became a commercial wedge.
• Customer Pull and Fast Product Validation - weekend experiment evidence that pushed the company away from the original path.