> ## Documentation Index
> Fetch the complete documentation index at: https://oxy.tech/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# World Model Graph, Secret Detail View, and On-Demand ELT Backfill

> An interactive graph of your semantic layer, a dedicated detail view for secrets, and on-demand backfill for date-windowed ELT sources.

<Icon icon="calendar" iconType="regular" size={16} /> **June 19, 2026** · v0.5.78, v0.5.79, v0.5.80, v0.5.81, v0.5.82, v0.5.83, v0.5.84

### New Features

#### World Model Graph

Explore your semantic layer as an interactive graph. The new **World Model** gives you a visual map of your entities, the measures defined on them, and how those measures flow across your entity hierarchy — reachable from the Globe icon in the rail.

* **Entity & measure graph** - Each entity renders as a node with its metric chips, plus clear badges for non-additive measures, fan-out risks, and broken measures, laid out with automatic orthogonal edge routing.
* **Instance drill-down** - Click an entity to open a searchable instance picker, then drill into a detail panel for any single instance.
* **Measure breakdown** - Expand a measure to see a driver tree of how it decomposes into the measures that feed it, shown inline on the node and in the detail panel — and only on measures that actually decompose.
* **Promotion paths** - Click an edge to see how a measure promotes or is induced as it moves across the entity hierarchy.

A `.world-model.yml` display config drives the entity and measure labels and an optional instance allowlist.

#### Secret Detail View

The unified secrets table now opens a dedicated detail view for each secret, gathering everything about a single secret into one focused dialog instead of spreading it across table columns.

* **Full secret at a glance** - See the value (reveal or copy it), where it comes from, its config reference, and when it was last updated — all in one place.
* **Source clarity** - Each secret is clearly labeled by source, so you can tell a database-stored secret apart from an environment variable at a glance.
* **Inline actions** - Edit, delete, or add an override for a secret directly from its detail view.

#### On-Demand ELT Backfill

Re-pull a bounded date range for date-windowed ELT sources (Toast and QuickBooks) without re-running an entire pipeline — useful for correcting or re-ingesting a specific historical window.

* **Bounded date-window replay** - Trigger a backfill for an explicit `from`/`to` range; the pipeline pulls just that window and merges results idempotently on key, so re-running the same window is safe.
* **Safe alongside live syncs** - During a backfill the source freezes its incremental cursor, so a bounded replay won't disturb the state of the ongoing incremental sync.
* **Streamed progress and run history** - Backfill runs flow through the normal pipeline coordinator and stream progress over live updates, with the requested window recorded in run history for later inspection.

#### Org Subdomain Routing

Select customer orgs can now be served on a clean, branded subdomain that opens straight into their workspace.

* **Branded org subdomain** - An org can be reached at `<org-slug>.oxygen-hq.com`, which boots the product pre-scoped to the org's default project and skips the org and workspace picker.
* **Custom apps on the same host** - The org's custom apps are served under the same domain at a tidy `/a/<slug>/` path.
* **Single, centralized sign-in** - Login and signup stay centralized on the main site, so authentication carries seamlessly between the subdomain and the main product.

Subdomains are enabled per org by the Oxy team; org owners see a read-only status line for their subdomain in settings.

#### Refreshed App Frame: Top Bar, Docked Ask, and Chat

Building on the apps-first HQ, Oxy now wraps every screen in a consistent app frame and turns Ask into a docked side panel you can keep open while you work.

* **Universal top bar** - A top bar now spans every page with a `Workspace / Page` breadcrumb (the workspace name links back to HQ), a timezone-aware clock that follows your default agent's configured timezone, a system-status indicator, and an "Ask Oxygen" button.
* **Docked Ask panel** - The floating Ask pill becomes a right-side panel that compacts the page beside it instead of floating over it. Resize it from its left edge, collapse it without losing your place, and reopen it with ⌘K — your conversation stays put.
* **Dedicated Chat page** - The rail's "Chat" button (formerly "Threads") opens a clean, ChatGPT-style composer with your workspace's recent threads listed below.
* **Searchable thread history** - The Chat page and the Ask panel now include a server-side search over your past threads — matching on both the title and the original question — with a "Show more" button to page through older conversations.
* **Simpler navigation rail** - The icon rail is reordered to **Home · Chat · Apps**, and the workspace's Oxygen Factory mark now appears as the rail logo in both light and dark themes.

***

### Platform Improvements

#### Reliability

* **Graceful Developer Portal restarts** - When the Developer Portal (Oxygen Factory) is briefly unavailable — for example during a restart — data apps, thread charts, and analytics runs now show a calm "restarting, resuming shortly" panel with a Retry option instead of a raw error. Your app sidebar, procedures, and semantic views keep rendering throughout, so navigation stays intact while the runtime comes back.
* **Bounded on-disk logs** - Self-hosted deployments no longer accumulate an unbounded `oxy.log` file. Local installs now rotate the log daily and keep the last seven days, while cloud deployments rely entirely on container log aggregation and skip the on-disk file altogether — preventing the slow disk fill-up that could destabilize long-running instances.
* **Cleaner Developer Portal git workflow** - Editing on a protected branch no longer leaves your `main` branch perpetually showing changes. Oxy now keeps its background working copies out of `main`'s git status, so `main` stays clean after you start an edit, pulling the latest no longer forces you to commit stray changes first, and the "Discard all changes" and "Reset to commit" actions work correctly on `main` again.

#### Settings

* **Airhouse version badge** - The Airhouse settings section now displays the running Airhouse server's version, so you can confirm at a glance which build your deployment is connected to. The badge stays out of the way when Airhouse isn't configured.

#### Semantic Layer

* **Branch-aware semantic previews** - View and topic previews in the Developer Portal now load from the branch you currently have checked out, so previewing a `.view.yml` or `.topic.yml` reflects your in-progress branch edits instead of always showing the `main` version.

#### ELT Performance

* **Faster, more reliable Toast order ingestion** - Toast `orders` ingestion now appends and compacts instead of running a costly per-row merge, eliminating the out-of-memory failures and timeouts that hit large order tables at millions of rows. A latest-wins compaction runs at the end of each load, so order data is fresh the moment the pull finishes.

#### Airhouse

* **Reliable high-volume ingestion** - High-frequency writes into Airhouse now reuse a persistent, per-workspace connection and batch rows into multi-row inserts instead of opening a fresh connection on every write. This eliminates the connection churn that could exhaust Airhouse warehouse memory under sustained ingestion, keeping streaming pipelines stable at scale.
* **Lower memory on large query results** - Running large analytical queries against Airhouse no longer materializes the full result into a temporary table before returning it, reducing memory spikes when working with big result sets.
* **Stable analytics under load** - Chat analytics, Data App tasks, and the SQL IDE now share a small pool of reused connections to Airhouse instead of opening a fresh connection for every query. This caps how many concurrent sessions a workspace can spin up, preventing the connection storms that could exhaust Airhouse server memory under heavy querying — and keeping unrelated ELT work from getting starved or timing out when analytics traffic spikes.
* **Coalesced high-frequency ingestion** - High-volume edge telemetry is now buffered in memory and written to Airhouse as one batched insert per stream per window, rather than committing every individual incoming request. Collapsing a flood of tiny writes into a single periodic write relieves the metadata churn and compaction pressure that destabilized Airhouse under sustained ingestion, keeping high-throughput streams stable.
