Rowset Weekly — Issue #3

Period: July 13–19, 2026 · Commits: 34

Repository: LVTD-LLC/rowset

If you run Rowset on your own infrastructure, build agents on top of it, or rely on its analytics pipeline, this week's updates are for you. We shipped backup/restore for self-hosting, made agent capability discovery topic-selective, and overhauled campaign attribution — plus a new agent benchmark harness and two in-depth guides.

Self-Hosting: Backup, Restore, Preflight, and Release Automation

Several foundational pieces landed to make self-hosted deployments production-ready.

• Backup and restore shipped (+1,535/−101) — The centerpiece: a self-hosting backup/restore workflow with isolated restore drill backups.
• Coherent immutable releases (+566/−68) — Published releases are now immutable and reproducible.
• Preflight and doctor commands (+1,127/−26) — New preflight and doctor commands for environment validation.
• Release contract and daily train (+546/−41) — Enforced release bundle contracts with an automated daily release train.
• Tested sizing requirements (+809/−9) — Published reproducible sizing benchmarks with recorded results.
• Fixed minimum disk preflight (+537/−124) — Defined a cost-first minimum profile to fix a false-positive preflight failure.

Agent Interface & MCP Improvements

The agent interface and MCP server received several targeted enhancements.

• Topic-selective capability discovery (+722/−71) — Agents can now filter capability discovery by topic, with CLI exposure and documentation explaining progressive discovery.
• Permission-aware tool discovery (+520/−59) — MCP tools now respect user permissions before surfacing available actions.
• Progressive Rowset discovery (+265/−105) — Fixed the agent to use progressive discovery for Rowset-specific capabilities.
• Server safety instructions (+45/−1) — Added essential safety instructions to the MCP server.
• Agent collection pagination (+256/−183) — Bound pagination on the agent collection API endpoint.

Analytics & Measurement

A significant analytics push this week covering attribution, pageviews, and lifecycle events.

• Campaign attribution tracking (+796/−273) — New campaign attribution system with consent-gated first/current touch, checkout session propagation, and proper exception handling.
• Campaign attribution sync fix (RMA-004) (+789/−95) — Synchronized consent-gated attribution across PostHog people and backend conversion events.
• Privacy-safe pageviews (+751/−1) — New privacy-safe pageview capture mechanism.
• Anonymous identity preservation (+74/−9) — Fixed a bug where anonymous identities were being lost.
• Login & account deletion lifecycle events (+377/−2) — New tracking for login and account deletion events, kept privacy-safe.
• Public measurement contract (+79/−2) — Documented the public contract for analytics measurements.

Onboarding & Auth

• Activate users after agent setup (+299/−131) — Users are now activated after completing agent setup, with refined trial rewards links and broader setup context guidance.
• Require bearer API keys (+531/−626) — Authentication now mandates Bearer API keys.
• GitHub social buttons polish (+7/−4) — Polished the GitHub social login buttons.

Admin Dashboard

• Product health at a glance (+844/−239) — Replaced the inventory-heavy admin panel with a product-health-first overview. Includes HTMX period switching, global mutation timeline indexing, and proper connector loading.
• Remove mutation metrics (+149/−90) — Cleaned up mutation metrics from the dashboard and excluded staff-owned data.
• Disable Django admin in production (+61/−7) — Django admin is now disabled in production for security.

Blog & Content

• Comparison article redesign (+232/−29) — Comparison articles got a full visual redesign.
• Aligned article actions with comparison posts (+38/−3) — Blog article actions now match the comparison post UI pattern.

New Guides Published

• Human-in-the-Loop AI Agents: A Practical Workflow (+438/−1) — A practical guide to building human-in-the-loop agent workflows (by Greg).
• AI Agent Audit Trail: What to Log and How to Build It (+365/−2) — Comprehensive guide on agent audit trail design (by Greg).

Datasets, CLI & Evaluation

• Multi-choice filters for datasets (+254/−39) — Dataset filtering now supports multi-choice selection.
• Agent benchmark harness (+1,478/−1) — New evaluation harness for benchmarking agent interfaces.
• Nested help and compact JSON output (+459/−90) — CLI now supports nested help commands and compact JSON output mode.
• Colorization enabled by default (+38/−1) — Profile colorization is now on by default.

SEO & Infrastructure

• Prevent self-hosted search indexing (+216/−49) — Self-hosted instances now block search engine indexing.
• Greptile review depth pinned (+59/−0) — Pinned the Greptile code review depth setting.
• Clean context-budget tests for MCP (+70/−8) — New tests enforce clean context-budget checks in the MCP layer.

That's 34 commits across 7 days. See you next week.