Journal
Blog
Engineering notes, product announcements, and practical guides from the Limerence team.
14 stories
A Virtual Filesystem in PostgreSQL
How Limerence gives every agent a POSIX-shaped tree by composing a two-table key-value store with scope-prefixing and mount-routing wrappers.
BigQuery, Meet Natural Language
Point an agent at a BigQuery warehouse — including cross-project public datasets — and ask questions in plain English without hand-qualifying project IDs or paying for queries that were never going to parse.
The Text-to-SQL Accuracy Playbook
The pipeline, invariants, and recovery loop that move text-to-SQL accuracy in Limerence — and the gaps that are still prose, not code.
Schedule Reports, Not Meetings
Turn a prompt you keep retyping into a self-running, browsable analysis with a run history you can scroll back through.
Automations That Schedule Themselves
A Limerence automation splits its lifecycle across two pg-boss queues so cron ticks never block on LLM latency and each queue carries the policy its workload actually needs.
BYO Keys Isn't a Feature, It's a Boundary
Whoever signs the LLM request is the party that sees the prompts — which is why text-to-SQL keys belong on your side of the wire, not the vendor's.
Defense in Depth: Keeping Read-Only Really Read-Only
Read-only AI database access is not one switch. Limerence layers four independent guards — scope, SQL validation, session locks, file protections — around your data.
When AI Writes HTML: The Interactive Elements Protocol
Streaming kebab-case HTML from an LLM into a live React tree works because five enforcement sites along the token path turn prose into a typed contract.
One Adapter Layer, Seven Data Source Types
How Limerence routes Postgres, MySQL, MariaDB, SQL Server, SQLite, BigQuery, and spreadsheet uploads through one shared query pipeline.
Ask Your Database from Slack
How Limerence's Slack and Google Chat integrations let teams query data without leaving their workflow.
Teaching Agents Your Business Language
How Limerence turns domain jargon into typed instruction fragments — authored by hand or harvested from schema analysis — and loads them into agent context at query time.
Your Clients Want AI in Their App — Not Another Login
Why embedded and workflow-native AI beats shipping one more standalone tool your customers have to remember to open.
Durable Streaming for AI Chats
How Limerence makes AI chat streaming durable with persisted chunks, replay-on-reconnect, cancellation, and startup recovery.
The Agent Has Questions For You
How we built AskUserQuestion — a tool that lets AI agents pause, ask structured questions, and learn from your answers.
Stay updated
New writing, shipped quietly.
Get an email when we publish something worth reading.