April 12, 2026
Postgres, Neon, and Drizzle
Serverless-friendly Postgres and a typed ORM; boring data paths keep the agent layer honest.
Field notes
Short memos on how lmkgpt is built, each tied to a primary, citable reference. No ranking lists; no “ultimate guide” packaging.
April 11, 2026
Specialists and capability packs
Not every subtask should look like a default chat; routing sends work to the right handler.
April 10, 2026
Mission knowledge and vector search
Structured facts and embeddings sit in Postgres so new missions can find prior work.
April 9, 2026
Content moderation before execution
Policy checks run on prompts before the orchestrator starts expensive work.
April 8, 2026
Trial rules and access control
Free runs are a product decision; throttling and abuse protection are a systems decision.
April 7, 2026
Auth, sessions, and guest identity
Signed-in users and guests share the core mission path with different persistence and limits.
April 6, 2026
Trust and full prompt transparency
Plans, system prompts, objectives, and raw outputs are visible; synthesis sits beside them.
April 5, 2026
Multi-provider routing and per-agent failover
Claude and GPT in one stack; one worker can fall back without taking down the run.
April 4, 2026
API-first orchestration
The same run users trigger in the app is a POST to an orchestration contract.
April 3, 2026
Real-time progress with server-sent events
One-way HTTP streaming for token and status updates, without websockets in the default path.
April 2, 2026
Parallel sub-agents and fault isolation
Why we fan out with settled promises, and what that buys when one worker fails.
April 1, 2026
From prompt to plan to a single answer
Why the coordinator pattern wins in production, and how we apply it end to end.