We write like humans who ship code. Field notes from production—SQL + Python wiring, prompt design that doesn’t break at 3 a.m., and deep-learning explainers your future self will thank you for. New agents. Honest benchmarks. No hype.
Curated drops: what we’re building, what launched, and the research that actually moved our roadmap.
We rewired discovery to use a lightweight SQL memory so the agent “remembers” prospects it scored last week. Fewer duplicates, better outreach, and a clean dashboard. See design notes.
Posted 2025-09-21
Tool-first routing, a “don’t guess” rule, and guardrails that reduced escalations by 18% across pilots. Plus our favorite evals for “did the agent actually help?” Read.
Posted 2025-09-18
A peek at what’s baking. If it ships here, it’s because a real business asked for it.
Answers phone + chat, qualifies leads, and books to your calendar. Escalates to a human when confidence drops. Built for service businesses.
Status: private pilots • Get on the list
An operator in your inbox: triages emails, drafts replies with context, opens tickets, nudges teammates, and closes loops. SQL memory + tool use by default.
Status: prototype • Partner with us
Lightweight data-gathering agent for market maps. Scrapes responsibly, normalizes, stores, and summarizes changes week-over-week.
Status: research • Get updates
L1 support that knows your docs and past threads. Clear handover to humans; everything logged with reasoning traces for audits.
Status: private pilots • See SaaS lineup
We like code that pays the bills. Expect readable math, small ablations, and stories from production—failures included.
Why action beats chatter. We wire an agent with function-calling, confidence bands, and a “don’t guess” rule. Includes Python + SQL starter snippets and eval harness.
Read the guide • 12 min
Vector DBs are great, but for many workflows a compact SQL memory wins on speed, cost, and control. We show schemas, pruning, and privacy.
Schemas & code • 9 min
Bookmark these—living references we update as the stack evolves.
Folder structure, `.env` tips, SQL memory schema, and sample evals for tool-using agents.
Top papers & posts on agentic systems, function-calling, toolformer ideas, and eval methodologies.
How we handle PII, model logging, and red-team prompts. Practical checklists for SMB deployments.
Public docs for agent endpoints and webhooks. Request access if you want to integrate early.
Partnerships, pilots, or press—say hi. If you’re building in the open, we love that too.
Or email directly: imran@xaiagent.ai
