Changelog

What shipped, when

A running record of features, fixes, and improvements to Fenlo.

August 2026

·v1.10
Feature

Zero-result search analytics for admins

Workspace admins can now see exactly what their team searches for but can't find.

  • A knowledge-gaps view in the usage dashboard: a daily trend of searches versus zero-result searches.
  • A ranked list of the most frequent queries your team looks for but can't find answers to, so you know what to add to your knowledge base.
·v1.9
Feature

Dollar-based spending limits and plan-tier reliability

Rate limits moved from raw token counts to real spend, alongside a cleanup of how plan tiers are resolved and enforced.

  • Workspace, user, and group limits are now a dollar spend cap over a time window instead of a raw token count, measured consistently across every model.
  • Admins can see their plan budget and balance while setting a limit.
  • The plan-tier UI self-corrects when an action needs a higher plan, instead of showing stale access and failing on submit.
  • Removed a dead per-tenant plan-tier cache; plan tier now resolves directly from the workspace's billing record.
  • Fixed the operator plan-grant tool and closed a loophole that allowed open-ended comped grants with no expiry.
  • Fixed paid-plan admin features (Groups, Standard Answers, Service Accounts) being wrongly blocked before the request's workspace was identified.
·v1.8
Feature

Enterprise upgrade requests and consistent plan-tier gating

Admins can start an Enterprise conversation from inside the app, backed by a single source of truth for what each plan unlocks.

  • Request an Enterprise upgrade directly from Settings > Billing — share seat count, timeline, and the capabilities you need, such as SSO, SCIM, and audit logs.
  • Consolidated plan-tier gating into a single backend registry that both the API and the UI read from, so a feature can no longer be gated differently in different places.
  • Cleaned up the Groups page now that group-based permissions are live and self-explanatory.
  • Startup reliability fix so the new Enterprise-upgrade request table initializes cleanly.
·v1.7
Feature

Configurable permissions per group

Groups became the place to control what members can do across the workspace.

  • Groups now have their own configurable permissions, covering agents, connectors, document sets, language models, actions, and more.
  • Manage them by opening a group from Settings > Groups.
·v1.6
Feature

Ollama support and a streamlined LLM setup flow

Bring your own self-hosted model, set up through a clearer language-model configuration flow.

  • Connect a self-hosted Ollama model as an LLM provider right from onboarding and workspace settings.
  • Redesigned language-model setup with a clearer provider list, showing Fenlo's built-in model alongside your own providers.
  • Existing deployments now auto-provision the group-based permissions schema on startup, so permission checks resolve correctly for previously-created workspaces.
·v1.5
Feature

Knowledge-graph-aware search (opt-in)

Search can now lean on extracted knowledge-graph entities to better surface the people, accounts, and projects a query is about.

  • Optional knowledge-graph re-ranking toward the people, accounts, and projects a query is about.
  • Off by default; when enabled it only re-orders results and falls back to standard search on any failure, so existing search behavior is unchanged unless a workspace turns it on.
·v1.4
Feature

Self-checking search retrieval and reliability guarantees

Retrieval learned to check its own work, backed by regression tests that keep chat citations honest on every release.

  • Self-checking retrieval: search grades whether it found enough to answer and re-searches with reworded queries when it hasn't, saying so when the knowledge isn't there.
  • Per-channel regression tests and a post-deploy canary verify web, Discord, and Slack chat still retrieve and cite connector documents on every release.
  • Blob/S3 connectors now reconcile source-side deletions during pruning, and a failed connector-deletion step no longer leaves the connector stuck before retrying.
  • Secondary search LLM calls (query rephrasing, keyword expansion) are bounded by a timeout and fail open instead of hanging a search request.
  • File Reader enabled for .xlsx/.csv uploads in standard deployments.
·v1.3
Feature

AI search summaries, response modes, usage credits, and Pulse

Search and chat got smarter and more controllable, with new response modes, usage-based cost accounting, and the first proactive insights.

  • AI-generated search summaries, with a filter to narrow results by connected app.
  • Redesigned, email-first two-panel onboarding.
  • Smarter search: query-aware ranking, fail-safe spell correction, and an app-wide keyboard shortcut system.
  • Fast / Thinking / Deep Research response modes with automatic routing.
  • Structured clarifying questions for ambiguous queries.
  • MCP Marketplace for one-click MCP server setup.
  • Rolling-window usage-credit accounting with flexible top-ups and per-tenant cost controls for contextual retrieval.
  • Default agent and project templates.
  • Pulse — proactive, relevant insights below the chat input, built on a new memory foundation (gradual rollout).

July 2026

·v1.2
Feature

Managed models, a Finder-style knowledge picker, voice, images, and people search

A large release that expanded what Fenlo can do — managed AI models, richer knowledge browsing, voice and image generation, and finding the right person — alongside a broad visual refresh and settings overhaul.

  • Fenlo-managed AI provider with free and premium model tiers under monthly cost caps.
  • Finder-style knowledge picker with both a flat browse view and a folder-tree view, now including files from your own personally-connected accounts.
  • AI image generation using your already-connected model providers via OpenRouter.
  • Voice input (speech-to-text) via OpenRouter, later moved to a dedicated transcription API for quality.
  • People / expertise search — find who on your team knows about a topic, based on where they're active.
  • Company announcements — admins can post org-wide announcement banners.
  • Personal connectors — connect your own accounts, starting with Slack, without waiting on a workspace admin.
  • Visual refresh with a calmer, more restrained design language, and a unified login/signup flow.
  • Keyboard accessibility with visible focus rings, plus toast notifications and component polish.
  • Consolidated, modal-based connector setup and reorganized settings, retiring the legacy admin console.
  • Hardened multi-tenant isolation across voice connections and background tools, plus OAuth and error-leak security fixes.
  • Reliability fixes: agents now search only their attached knowledge, stuck new-chat streaming resolved, and Google sign-in fixed on split web/API domains.
·v1.1
Improvement

Connector admin cleanup and onboarding fixes

A round of polish on the connector admin experience and onboarding reliability.

  • Cleaned up the connector and indexing admin UI to match the design system.
  • Fixed a broken checkout flow during onboarding.
  • Faster loading of built-in LLM model options.
·v1.0
Improvement

Launch hardening: connectors, security, self-hosted database

The first wave of post-launch work focused on making connector setup, indexing, and the platform's security foundations dependable.

  • Smoother GitHub connector setup — fixed OAuth visibility, a nested form, and a slow setup query.
  • Connector credential-decryption failures now fail gracefully instead of surfacing raw errors.
  • Security and reliability hardening across auth, chat, and file handling: session and lockout handling, decompression-bomb protection on uploads, and safer error surfaces.
  • GitHub source-code indexing via AI-generated code summaries.
  • Honest connector indexing status, with a discoverable re-index button and infrastructure stability fixes.
  • Quieter, less noisy error monitoring on the LLM provider settings endpoint.
  • Migrated off a free-tier hosted database to a self-hosted Postgres instance included in the deployment.