Local project workspaces
Create, import, list, search, update, archive, and remove project records while keeping the repository files where they already live.
FORGE / FEATURE REFERENCE
Forge is a local-first workspace for software work. This reference covers its surfaces, operations, boundaries, and integrations.
Workspace foundations
Create, import, list, search, update, archive, and remove project records while keeping the repository files where they already live.
Inspect repositories, dependencies, symbols, architecture, onboarding context, branches, diffs, and project activity from one workspace.
Review background work, task status, recent activity, and recoverable async operations while keeping the project thread.
Search commands, projects, sources, tasks, and product surfaces from a keyboard-first command palette that stays closed until you open it.
Choose privacy mode and model provider, review desktop permissions, inspect connection status, and keep local configuration visible.
A persistent onboarding guide introduces project creation, repository connection, and the first analysis, with progress that can be resumed from Launch.
Neuron
Ask questions using approved project context, with provider selection and invocation boundaries that respect the active workspace.
Index local resources and keep source boundaries, freshness, citations, and provenance attached to the material Neuron uses.
Search with lexical, vector, hybrid, and reranked retrieval strategies so useful answers can be found across the project's connected resources.
Add, list, query, and remove local knowledge sources without pulling in unrelated material.
Review and manage accepted memories so useful context can persist deliberately instead of becoming invisible model state.
Save, list, and reopen local sessions so questions and answers remain attached to the project they explain.
Pulse
Write local journal entries and keep ideas, observations, and decisions connected to the moment they came from.
Capture active-app context, focused windows, selected text, clipboard content, and screen context only when the required permission is granted.
Manage profiles that shape how Forge behaves across different projects, work modes, and contexts.
Start known prompts and workspace actions from explicit keyboard launchers instead of rebuilding the same setup each time.
Parse voice command text into an inspectable action without running anything silently or automatically.
Define bounded automation rules with visible boundaries, local persistence, and recoverable outcomes.
Keep clipboard filtering and sensitive-content rejection enabled so capture does not silently absorb secrets or private material.
Browser Companion
Capture a page or selected text while preserving its source URL and provenance for later review.
Save web context for later with its source attached to the project thread.
Keep a visual record of the page context that informed a decision, review, or bug report.
Inspect the structure behind a page so interface context can be understood rather than copied as an opaque screenshot.
Review page accessibility findings with the browser connection and its permission boundary made explicit.
Review screenshots, inspect frontend structure, and extract design tokens that can return to the project with their source intact.
Create traceable bug reports from browser context and preserve the evidence needed to understand the issue.
Detect meeting context, ingest transcripts, summarize them, extract action items, and link the result back to a project.
Foundry
Define agents and tools with declared capabilities instead of treating automation as an opaque prompt with unlimited reach.
Compose repeatable workflow definitions that turn a known project move into a path other people and future-you can inspect.
Keep unsafe tools behind explicit approval and capability checks before an action can cross a sensitive boundary.
Run steps with retry behavior, failures in view, and recovery paths that do not hide partial work.
Bound workflow runs with budgets, progress, cancellation, and clear outcomes for long-running work.
Visualize workflow progress and retain events that explain what ran, what was approved, and what happened next.
Architect and Docs
Build editable graphs of services, packages, databases, external systems, and decisions, then connect them to the project story.
Record ADRs and rationale so the decision remains clear after the diagram and the original conversation have moved on.
Preview generated project starting points before accepting them into the local workspace.
Create and edit README, architecture, API, onboarding, and runbook drafts while keeping generated material editable.
Review whether documentation still reflects its sources before exporting it as an artifact.
Export diagrams, evidence packages, case studies, and career-oriented project summaries when the underlying work is ready to be shared.
Git
Read the change in context, collect evidence, and make the implications visible before applying anything.
Preview mutations, keep destructive actions behind confirmation, and preserve rollback paths when changes are applied.
Discover project tests and interpret their results as part of the change review rather than as an afterthought.
Generate commits and pull requests from reviewed work, with the mutation boundary explicit.
Add review comments to pull requests through an approved GitHub connection when the project authorizes that write.
Quality
Run bounded accessibility, performance, hydration, bundle, and Web Vitals checks with findings that remain inspectable.
Scan for security issues and exposed secrets without turning a missing or partial result into a made-up score.
Evaluate retrieval datasets and relevance so Neuron quality can be improved with evidence.
Simulate Foundry planning, tool use, approvals, retries, budgets, cancellation, and failure recovery deterministically.
Compare browser, Electron, service, companion, and integration adapters for schema, permission, offline, and capability parity.
Review milestones, platform gates, readiness, and clear limitations in one local quality and release dashboard.
Plugins and integrations
Inspect plugin manifests and requested capabilities before enabling an extension.
Sign and verify plugin packages so unsigned or tampered packages do not activate by default.
Keep plugin actions isolated behind declared capabilities, permission approval, and a recoverable disable path.
Review the network scopes and operations an external service can use before a connection is configured.
Store integration credentials behind the secret boundary, revoke them explicitly, and surface rejected credentials as recoverable errors.
Use authorization-code flows with PKCE, inspect the scopes being requested, and retain enough consent history to understand when access was granted or revoked.
Receive provider events with signature verification, replay protection, cursors, and approval-aware processing instead of trusting an unverified callback.
Keep provider health, rate limits, retries, token refresh, and degraded states visible so an unavailable service does not become a silent failure.
Team
Use accounts only when the collaboration workflow needs them; local work does not depend on an account by default.
Opt into end-to-end encrypted sync when a project needs to move between trusted workspaces.
Keep personal and shared scopes visible, assign roles, and make the boundaries of collaboration clear.
Separate organization data, membership, and project access so a shared project never becomes a path into another tenant's context.
Invite people to an organization or project, change roles, revoke access, and keep those changes explicit and auditable.
Support SSO, SCIM provisioning, verified domains, MFA policy, and service accounts when an organization needs its identity system to remain authoritative.
Share selected knowledge, coordinate approvals, and preserve the audit activity around collaborative decisions.
Queue encrypted changes while offline, retry safely, acknowledge accepted records, paginate pulls, and preserve version information for recovery.
Surface synchronization conflicts with the competing versions and an explicit resolution path rather than silently overwriting context.
Enroll trusted devices, deliver project keys through the device boundary, and rotate keys when membership or device trust changes.
Preview remote deletion, apply retention rules and legal holds, track deletion jobs, and restore an approved copy without hiding its status.
Support sign-in, refresh, recovery, sign-out, and account deletion as explicit states, without making an account prerequisite for local work.
Launch and recovery
Understand what Forge can do, what is available in the current runtime, and which permissions a capability needs.
Create release metadata and checksums that make the artifact and its origin easier to verify.
Back up local state and export project artifacts without making network sync a requirement.
Collect recovery-oriented diagnostics while excluding prompts, source text, clipboard contents, secrets, and raw identifiers.
Keep the operational story documented so updates, recovery, and known limitations remain actionable.
Show when a capability is unavailable, why it is unavailable, and what the user can do next.
Expose API health, readiness, dependency status, correlation IDs, and audit events so support work can follow a request across the system.
Pair backups and exports with restore verification, disaster-recovery procedures, deployment rollback, and clear recovery ownership.
Define SLOs, alerts, incident response, provider setup checks, and staging gates before a hosted workflow is treated as ready.
Supported integrations
Forge currently defines configurable adapters for GitHub, Jira, Linear, Slack, Notion, Figma, Vercel, Supabase, Gmail, Calendar, and Obsidian. Each connection is optional, scope-limited, credential-bound, and protected by explicit network permission.
Read operations and write operations are distinguished. Writes require explicit confirmation, rate limits and retries are handled, and revoked credentials return a recoverable connection state.
When a backend is configured, authenticated clients can create, list, inspect, and select projects through an organization-scoped API without exposing local repository contents by default.