Alan's Auto Router is legit: 10/10 on my coding tasks. So I tried to beat it 👀
Problem: agents send every call to a giant model. Slow.
Pennywise: an RL-trained router that picks the right model per call
Same 10/10, 3.4× faster. Runs inside @tryAlanAI as a connector.
Today we're launching @tryAlanAI so developers can stop babysitting their coding agents.
We kept running into the same problem ourselves.
Even as agents got better at writing code, we were becoming glorified button-pushers for our own AI: starting every run, feeding them context, unblocking failures, and checking their work.
We built Alan to change that.
Dear @tryAlanAI, we need to talk about your router.
Pennywise, an RL-trained router, vs Auto Router on 10 coding tasks:
⚡ 3.4× faster (11 vs 37 min)
✅ both 10/10
npx -y pennywise-mcp
getpennywise.pages.dev
@prasad_pilla @lowkeyarhann @amannhq @sidtalesara @idleshubh
• 948 prompts × 3 open models = 2,844 graded answers
• Laya decision model, SFT then REINFORCE
• routes each call to small / medium / large
• checks agent tool calls, escalates when the fast model slips
Paper: getpennywise.pages.dev/paper…
Code: github.com/1sakshm/pennywise
#Openfire