Blog

Notes from the build

How SeldonFrame is built, why we made the architecture choices we did, and what we learn from dogfooding it on real customer operations.

May 7, 2026·SeldonFrame Team

Why we built SeldonFrame on MCP

Most AI startups wrap a single chatbot UI around an LLM. We took the opposite bet: expose the whole product as MCP tools and let Claude Code be the chrome. Here's why.

Read post
Coming soon·SeldonFrame TeamComing soon

How the eval gate works (and why agents need one)

A chatbot that hallucinates a price or claims it booked an appointment when it didn't isn't a bug — it's a real-money problem. The 8-scenario suite, the LLM-as-judge rubric, and the runtime regeneration that catches what the suite misses.

Coming soon·SeldonFrame TeamComing soon

BYOK is not a feature, it's the deal

Why SeldonFrame doesn't markup tokens, doesn't pool LLM access, and doesn't ration by tier. The economics work better when the customer owns the provider relationship.

Follow @seldonframe on 𝕏 for new posts, or join the Discord to talk to the team.