Mubit trains small models from your own agent API traffic to cut cost and latency
rohanpaul_ai · x · 2026-10-24
Aberdeen-based startup Mubit launched a service that reads the API calls your agents already send to OpenAI or Anthropic, spots repeated tasks, and trains small models you own from that real traffic.
- Narrow, repetitive work moves to your private small model; everything else still routes to the frontier model, cutting both cost and latency
- Models refresh every few minutes, per the company
- Integration is one line: swap the base URL and keep the same client code, or connect traces you already collect
- Two companies on Mubit end up with different small models, each shaped by its own agents' work
- The founders argue that remembering feedback via prompts is unpredictable, so they bake it into training instead
Launch comes with a promo: comment "SAPLING" within 48 hours for $100 credits, first month free for enterprises.
Related event: Mubit Turns Agent Traffic into Proprietary Small Models(2 posts)→
More from coding & agent
- Codex Computer Use fails to capture Safari on Mac mini M4 with SCStreamErrorDomain -3811/-3812 — darkprinceimmortal · 2026-10-24
- Heavy AI user: marketed automations are fragile, exhausting, and dumber than we admit — mrks86 · 2026-10-24
- Google's playbook: evaluating AI agents beyond the vibe check — bibryam · 2026-10-24
- Running CI weekly on dotfiles setup.sh inside a fresh Ubuntu container to catch breakage — 4310sy · 2026-10-24
- Codex CLI's hidden remote-control feature praised as better than Claude Code's — antirez · 2026-10-24
- PWAs can't use foldable screens on the Duo yet, but two Chrome flags unlock the APIs — mitsuhiko · 2026-10-24