📢

v4.10.0 — One AI budget, clearer Findings

A shared monthly AI allowance and focused Finding detail make usage, evidence, and resolution easier to understand.

Novidades

Fique por dentro das últimas atualizações e melhorias

A shared monthly AI allowance and focused Finding detail make usage, evidence, and resolution easier to understand.

Galileo 4.10.0 gives every organization one honest AI budget and makes Findings easier to inspect and act on. Usage stays visible from monthly allowance through individual generations, while focused previews and a canonical detail page keep the evidence behind each Finding close at hand.

New

  • One monthly AI allowance — generation, Retrieval, embeddings, and analysis now share the same provider-reported organization budget, with plan-aware limits and honest remaining or overage totals.
  • AI usage settings — admins and owners can inspect period history, generation-level consumption, usage outside generations, and configure caps for future analyses.
  • A focused home for every Finding — list previews lead with one divergence at a time, while a canonical detail page brings together status, evidence, work items, snooze controls, and every wording.

Improved

  • Safer background processing — Galileo now runs on Oban's latest PostgreSQL schema, including support for suspended jobs and a migration contract that protects fresh installations.
  • More readable Finding evidence — long excerpts stay compact in previews and detail views, with the original source or edge always one click away.

Fixed

  • GitHub permission links reach the right account — organization installations now open their organization settings instead of a personal route that returns 404.
  • AI usage remains stable while navigating periods — generation rows no longer disappear after a LiveView update, single-month badges do not create horizontal scrolling, and the cap action aligns with its input.
  • Resolved Finding previews keep their spacing — the featured divergence no longer touches the expanded-row border when lifecycle actions are absent.

Turn Findings into trackable work with provider write-back, lifecycle sync, snooze, saved views, evidence deep links, and clearer explanations.

Galileo 4.9.0 closes the loop between finding a problem and getting it resolved. Findings can now become provider-backed work, stay synchronized through their lifecycle, and reach the right people with the context they need to act.

New

  • Turn Findings into work — create linked GitHub issues or Notion comment threads from a Finding, with access-safe context and idempotent lifecycle synchronization.
  • Finding alerts — newly detected Findings reach eligible members in-app, with a concise email digest for each detection run.
  • Saved Finding views — keep named filter presets for the slices of Findings you return to most often.

Improved

  • A calmer triage queue — managers can snooze a Finding for 7, 30, or 90 days, add a note, and reveal snoozed work when needed.
  • Evidence that opens at the source — GitHub evidence links target the captured commit and line or heading, while Notion evidence links to the captured block.
  • Clearer explanations — Findings and Graph details share a “Why Galileo flagged this” rationale, generated in the organization’s configured language for new claims.
  • Safer navigation and actions — claim details, dismissals, and return links remain scoped to the current organization and its access policy.

Dismiss or restore individual Finding claims without changing lifecycle status, backed by a sharper Conflict and Drift precision wave.

Galileo 4.8.0 makes Findings easier to act on and harder to get wrong. Managers can dismiss or restore individual claims without changing the Finding's lifecycle, and a sharper precision wave cuts noisy Conflict and Drift publishes before they reach the list.

New

  • Claim-level triage — dismiss a false or acknowledged claim as a reversible suppression, without writing over the Finding's status. Fully dismissed rows stay out of the default list, and you can restore a claim when the judgment changes.
  • Gen-2 precision wave — Conflict and Drift prompts advance with tighter scope, same-referent, and verbatim-span rules, and a Prompt version bump alone re-analyzes the corpus in budgeted waves.
  • Impact bands on every Claim — each published disagreement carries a high / medium / low impact rating from the judge. The band is stored now and surfaces in the UI only after enough dismissal history exists to trust it.

Improved

  • Fewer noisy publishes — Conflict caps at three proposals per subject (matching Drift), and anti-pattern list items framed as things not to do are rejected before they become Findings.
  • Grounded siblings survive — on both semantic axes, one ungrounded quote no longer discards the correct claims that arrived with it.
  • Stronger evaluation gates — the offline Judge baseline was re-accepted after the precision wave, keeping Conflict and Drift from quietly regressing.

Fixed

  • Severity required before verdict filtering — a missing impact band can no longer slip through as an empty valid response on either axis.

Findings are now identified by the content they point at, so dismissals stick — backed by an offline evaluation harness that keeps Conflict and Drift precision from regressing.

Galileo 4.7.0 is about earning your trust in what the analysis publishes. A Finding is now identified by where its claim lives rather than by the exact words the model quoted, so dismissing one makes it stay dismissed. Behind that sits an offline evaluation harness that replays a labeled corpus against the real prompts and refuses to let precision or recall quietly regress.

New

  • A judge for the judge — a versioned corpus of labeled Conflict and Drift verdicts replays through the same prompts and validators production uses, and every change is compared against an accepted baseline. Precision and recall can no longer drift without someone noticing.
  • Real-time Analysis coverage — the Sources page updates Processing, Partial, and Complete states and generation changes live, instead of waiting for a reload.
  • Observability and spend visibility — every AI call is recorded with cache-aware token accounting, exported as metrics, and watched against service-level objectives that notify admins on breach.
  • Browser end-to-end coverage — a real browser now exercises the interface on every change, catching the interactive behavior that server-rendered tests cannot see.

Improved

  • Dismissing a Finding sticks — a claim is identified by the content it points at, not by the sentence fragment the model happened to quote, so the same issue no longer comes back reworded on the next analysis.
  • Fewer wasted analyses — a single fabricated quote used to discard the correct findings that came back with it; the grounded ones are now kept.
  • Rate limits pause analysis instead of sacrificing it — a provider throttling window now backs off and resumes rather than permanently degrading the pairs it touched.
  • Clearer coverage reporting — per-Source details explain what was analyzed, what was skipped, and why, in plain language.

Fixed

  • Interrupted analysis recovers on its own — deploys and restarts no longer strand or duplicate long-running reasoning work.
  • Coverage no longer reads Complete too early — a Source added to an existing generation reopens the semantic axes instead of inheriting a finished badge.
  • Deleting a Source no longer live-locks analysis — stale plans are pruned in the delete, so budget is not burned in a loop against a generation that no longer exists.
  • Export links survive delivery — long download URLs no longer overflow the notification field and lose the requester's only pointer to the export.

Sharper analysis with fewer false-positive Findings, a plan-aware LLM budget you can see, faster reasoning, and a more polished, localized workspace.

Galileo 4.6.0 makes your knowledge analysis both sharper and more honest: precision gates cut false-positive Findings, the LLM budget follows your plan and is visible as a live meter, and reasoning runs faster so Findings land sooner after a Sync. Along the way we smoothed dozens of rough edges across the org workspace — including a fully localized pricing page and a clearer analysis coverage sheet.

New

  • Precision gates for Findings — deterministic gates retire four production false-positive modes, so the conflicts, doc↔code drift, and stale-doc Findings you see are the ones worth acting on.
  • Analysis coverage details — a per-Source sheet shows what was ingested and how much of each axis was checked this cycle, now with plain-language section subtitles, a "Why some items were skipped" breakdown, and an info tooltip on every axis.

Improved

  • A budget you can read — the analysis LLM-call ceiling now follows your plan (Team 300, Scale 1000) and appears as a usage meter under Settings → Advanced, with a gentle nudge as you approach the limit. Reasoning throughput roughly doubles, so Findings arrive sooner after a Sync.
  • A more localized workspace — Sources, Findings, Graph chrome, and the Subscribe pricing cards now read correctly in Brazilian Portuguese instead of falling back to English.
  • A calmer Add-Source form — the Connection field no longer flashes an "is required" error before you've touched it, and switching Source types clears scope cleanly.

Fixed

  • No more stuck "Processing" — when a cycle's budget is exhausted with work still pending, the analysis plan now finishes as Partial with a clear reason instead of showing an endless Processing badge.
  • Honest cost accounting — every model response is metered against the budget, so what you see reflects real spend.

OAuth Connections, Source Access policy, Labels, Active Evidence, deep markdown docs linked to code symbols, and a Retrieval foundation that only cites what you can still see.

Galileo 4.5.0 makes Orgs feel like a real workplace graph: OAuth Connections, Source Access policy, Labels for slicing the view, deep markdown docs linked to real code symbols — and a grounded Retrieval foundation so answers (soon) can cite Evidence that is still in scope and still visible to you.

New

  • Connections — authorize GitHub App installs and Notion workspaces once via OAuth, then create Sources under them with the right scope (repo + ignore, or Notion roots). No pasted API tokens on Sources.
  • Access policy — keep a Source open to every Org member by default, or tighten by minimum Org Role and/or an allowlist. Visibility is enforced on Sources, graph, Findings, and Retrieval — manage rights stay separate.
  • Labels — tag Sources from the form (create-on-type) and filter Sources, Findings, and the dashboard by Label. Labels organize; they never grant access.
  • Active Evidence — when material leaves Source scope, Evidence is deactivated (history kept). Findings and Reasoning only read what is still Active.
  • Deep GitHub docs — every in-scope .md/.mdx (not just the README) becomes a document with bounded Content blocks, ready for Conflicts and Findings.
  • Code symbols + doc links — Sync extracts public Elixir and TypeScript/JS symbols, then deterministically links docs that name them. Stale Findings deepen when docs still mention symbols that disappeared.
  • Retrieval foundation — Active Content blocks are indexed for semantic search with grounded Evidence hits (Access policy enforced). Ask/MCP surfaces come next — this is the shared substrate.

Improved

  • Source delete cleanup — removing a Source prunes orphaned Nodes, Edges, and Findings so the Graph does not keep ghosts of deleted ingest.
  • Add Source modal — Cancel returns you to the Sources list cleanly, so opening Add Source again always works.

Connect Notion beside GitHub, auto-propose cross-source relationships, and surface conflicting claims with both Evidence trails side by side.

Galileo 4.4.0 connects Notion beside GitHub, proposes cross-source relationships automatically, and surfaces conflicting claims — so lying documentation shows up with both Evidence trails side by side.

New

  • Notion Connector — connect a Notion workspace as a Source; every page the integration can read lands as a document Artifact with bounded page-block Evidence.
  • Cross-source Edges — after Sync, Galileo proposes Notion→GitHub describes Edges (LLM suggests, deterministic grounding decides), so Findings can span both Sources.
  • Conflicting-claims Findings — when two Sources disagree about the same service, Galileo opens one Finding on that service with both grounded quotes side by side — never picking a winner for you.
  • Source edit — rename a Source or rotate its credential without deleting it. Identity stays immutable; a credential change is re-verified against the same system.
  • Duplicate-Source protection — connecting the same GitHub repo or Notion workspace twice is blocked at creation with an immediate form error.

Improved

  • Realistic README grounding — GitHub READMEs are stored as structured blocks (like Notion pages), so cross-source quotes match what you actually read — not raw markdown markup.
  • Notification delivery — invitation emails honor email opt-outs for existing users; bulk changelog announcements are safely re-runnable without double delivery.
  • Admin and org tables — Graph, Findings, and admin indexes share one DataTable driver for search, filters, and live refresh.

Fixed

  • Transferring ownership to yourself no longer leaves the Org without an owner.
  • Cross-source proposal calls have more headroom so thinking models are less likely to truncate mid-response.

Org-wide Findings triage, live-updating dashboard stats, real Team/Scale billing, and a product-true landing page.

Galileo 4.3.0 brings org-wide Findings triage, live-updating dashboard stats, real Team and Scale billing, and a landing page that reflects what actually ships.

New

  • Org-wide Findings page — triage every Finding in your Org from one read-only table: kind, status, affected Node, and when it was detected or resolved. Filter, sort, and paginate via the URL; open Findings appear first.
  • Live updates — the Graph explorer and Findings page refresh in place when your Org's knowledge graph or Findings change, without losing your place.
  • Live dashboard stats — the Org dashboard shows live counts for Sources, Nodes, and Open Findings, each linking straight to the relevant page.
  • Plan entitlements — connected Sources are now metered by plan: Beta (2), Team (10), Scale (unlimited). The limit is enforced server-side, with an upgrade prompt when you hit the cap.
  • Brazilian Portuguese locale — Português (Brasil) is now available in the language switcher (pt_BR). Catalogs ship untranslated for now, falling back to English.

Improved

  • Honest landing page — placeholder copy, fabricated testimonials, and invented stats are gone. The hero shows a real knowledge-graph screenshot; shipped vs. roadmap capabilities are clearly labeled.
  • Real billing plans — Team ($99/mo or $990/yr, 10 Sources) and Scale ($499/mo or $4,990/yr, unlimited Sources), Org-billed with unlimited seats.
  • Org data export gated to Scale — full Org export (Sources, graph, Findings) is a Scale entitlement; free and Team Orgs see an upgrade message instead.

Fixed

  • DataTable pages no longer crash on tampered filter URLs — invalid enum values degrade gracefully to the default unfiltered view.

Galileo now flags documentation that has fallen behind your code, adds deep-dive graph pages, and ships production upload fixes.

Version 4.2.1 turns the knowledge graph into something you can explore end to end — and starts surfacing stale documentation before it becomes a surprise.

New

  • Stale documentation Findings — after a GitHub Source syncs, Galileo checks whether a README has fallen materially behind the code it describes. Open Findings appear on the document's detail page in plain language.
  • Graph explorer — browse every Node in your Org from a sortable, paginated index with filters that live in the URL.
  • Node and Edge detail pages — click through to properties, connections, and a provenance timeline for each piece of knowledge.
  • GitHub freshness signals — READMEs and root-level files now carry last-commit dates, giving drift detection real timestamps to work with.
  • Findings in Org export — your full Org export now includes Findings alongside Sources and the knowledge graph.

Improved

  • Findings resolve automatically — when documentation catches up or the underlying basis goes away, Galileo marks the Finding resolved without manual cleanup.
  • Production file uploads — cover images and admin file uploads work against S3-compatible storage in production.

Fixed

  • Admin file uploads no longer crash when a presign or upload is cancelled mid-flow.
  • Changelog titles with accents and em dashes save correctly.

Connect GitHub repos as Sources, sync them into a provenance-tracked knowledge graph, and export your Org's data on demand.

Galileo's core loop lands in 4.1.0: connect a Source, watch it sync, and explore the knowledge graph that grows from it — every Node traced back to where it came from.

New

  • Knowledge Sources — connect a GitHub repo per Org, watch sync status update live, and keep API tokens encrypted at rest.
  • GitHub Connector — syncs your repository into the knowledge graph: the repo becomes a service, its README a linked document, recent commits mapped with authors, and root-level files captured as code artifacts.
  • Reliable sync — re-sync converges without duplicates; rate limits retry automatically; clear, credential-free errors when a token lacks access.
  • Knowledge graph — Artifacts and Concepts linked by typed relationships, with append-only Evidence tying every Node and Edge back to its Source.
  • Org data export — Org owners and admins can request a full JSON export of Sources (credentials redacted), the knowledge graph, and provenance Evidence. A download link arrives by notification when it's ready.

What's next

More connectors (docs sites, Notion, Jira, Slack, and others) and cross-source reasoning are on the roadmap. GitHub is the first live connector — the pipeline every future Source will plug into.