Forge mode
Forge mode is where you build on the platform. Three clusters of work live here: working with data, running code, and extending the platform. Forge opens on a Pulse dashboard that surfaces the operational signal a developer most often wants — your collection count, deployed function count, installed extensions, pending migrations, recent invocation failures, subscription stream health, storage usage, and any usage-cap events from the last day. Click any widget to drill into the surface underneath, with a pre-applied filter matching the widget's scope. A "Back to Workshop" breadcrumb at the top of the drilled-down surface brings you back to Pulse without losing where you were.
Working with data
The Data cluster covers everything to do with the data plane — defining collections, evolving schemas, browsing records, querying, storing files, and watching real-time subscriptions.
Collections lists every collection in your project — the table-like containers your records live in. Browse, see row counts and last-migration timestamps, click into a collection to explore the records inside. There's no "create collection" button; collections come into existence through migrations.
Clicking into a collection takes you to a per-collection detail view with four sub-routes — Documents (the default; a query terminal scoped to this collection), Schema (read-only column types, indexes, foreign keys, policy rules), Indexes (every index, with a Create-index action that opens a pre-populated migration), and Subscriptions (live subscriptions filtered to this collection).
Migrations is where DDL changes go. Author a forward + backward migration pair, validate the SQL, dry-run against the current schema snapshot, commit atomically. Failed migrations roll back cleanly. The migration list is the canonical history of how your schema got to where it is.
Storage covers buckets, objects, signed URLs, image transforms, versioning, and lifecycle policies. The SDK transparently switches to TUS for files ≥ 5 MB; everything below that is a regular upload.
Subscriptions is a live view of every active and recently-closed real-time subscription on the project. Filter by collection or by state; inspect the recent event log; watch the per-subscription event rate. The surface to reach for when you're debugging "is my client actually getting events?" questions.
Metadata is a read-only view of the platform's per-record metadata — shard ID, created-at, updated-at — surfaced per-collection so you can do shard-distribution-aware planning or scope a time-window investigation without writing the SQL.
Queries is read-only raw SQL against your project's data — a four-pane layout (schema browser, SQL editor, parameters, results) and a Run button. Reach for it when you're debugging at the infrastructure layer and don't want the composability overhead of authoring a Studio Data Explorer query. There's a "Save to Studio" action that promotes the query into a named, versioned, composable artifact if you decide you want one later. Queries never mutate data, for any role: changes go through the governed Document API.
Running code
The Compute cluster is where your runtime plane lives — functions, the LLM library, invocation usage, and project-level compute config.
Functions is the function registry — deploy, edit, wire triggers, inspect logs, replay failed invocations. Triggers cover collection-write events, scheduled cron, HTTP endpoints, and extension-emitted events. Functions are versioned; deployment is a commit-and-apply path rather than a hot-edit.
LLMs lists the curated-library models your functions can call. You will find models such as phi-3-mini and mistral-7b. LLM identifiers are what your functions and any AI-agent extension's manifest reference.
Usage shows per-function and aggregate invocation metering — call counts, execution time, billing dimensions, a time-series visualisation, filters by function and trigger source and time range. The aggregate feeds your project's billing.
Settings is per-project compute limits — max concurrency, execution timeout, memory cap, GPU governed-operation routing preferences for sovereign deployments, and the default function tier (platform for Bun + WebAssembly, or customer for Firecracker microVMs).
Extending the platform
Extensions at the foot of the sidebar lists every extension installed on your project, with a path to the Extensions Hub. Install or enable an extension, review its manifest permissions, configure it, or remove it. The four Trailblazer-class extensions (Paperless, Context, Lookout, Gate Studio) live here alongside any Anchor or Circuit extensions your team has installed.
Read-only access
A PROJECT_VIEWER can browse every Forge surface — collections, migrations, storage, subscriptions, functions, LLMs, usage metrics, installed extensions — but the mutation actions (Edit, Delete, Deploy, Toggle trigger, Run write-mode query) are disabled with a tooltip explaining why. Useful for compliance reviewers and read-only operators.
See also
- Orientation — the five modes at a glance.
- Admin — for operator work.
- Vault — for encryption custody operations.
- Studio — for authoring views, forms, workflows, and the rest.
- API → Data — the SDK and HTTP surface behind the Data cluster.
- API → Compute — the SDK and HTTP surface behind Compute.
- API → Extensions — the framework contract behind Extensions.
- Specifications → Billing — how usage and invocation metering roll up into your bill.