Platform Documentation

Complete reference for Forge — the AI-powered semiconductor test automation platform by deepsilabs.

Overview

Forge is a collaborative workspace for semiconductor design teams. It combines project organisation, an AI agent that synthesises test code from natural-language descriptions, file storage for design documents, and a full test-execution tracking pipeline — from initial task staging through physical hardware execution to results analytics.

  • Multi-workspace — switch between isolated team environments
  • Role-based access — Admin, User, Observer, Technician
  • AI-driven synthesis — describe tests in plain language, get production code
  • End-to-end tracking — staged → queued → executing → completed
  • Real-time streaming — agent responses and tool calls streamed live via SSE

Dashboard

The dashboard gives a high-level view of workspace health at a glance.

Metric Cards

  • Active Tests — count of tasks in staged, queued, synthesizing, needs_review, and execute states; includes an "awaiting review" sub-count
  • Success Rate — completed ÷ (completed + failed), shown as a percentage
  • Projects — total project count and aggregate task count

Weekly Analytics Chart

An 8-week area chart showing Tests Submitted and Tests Completed per week, aligned to Monday boundaries. Data is cached for 60 seconds and refreshed silently in the background so switching tabs does not cause a reload flash.

Projects

Projects are the top-level containers for test campaigns. Each project holds its own file store, chat history, and task list.

Projects List

  • Create — name + optional description; Enter key submits
  • Rename — click the pencil icon on any project card
  • Delete — admin-only; cascades to tasks, files, and chat history with a confirmation warning
  • Card layout — responsive 1 → 2 → 3 column grid with hover action buttons

Project Detail — Two-Panel View

Opening a project splits the screen into two resizable panels:

  • Left panel — Chat & Upload — AI chat interface with drag-and-drop file upload
  • Right panel — Files & Tasks — file browser with preview and the project task list

Tasks

Tasks represent individual test items — from AI-synthesised code generation jobs to physical execution runs. The Tasks page provides a cross-project view of every task in the workspace.

Task List

  • Columns — task name, project, status, priority, assigned user, created date
  • Filter by status — any single status or show all
  • Search — filter by task name substring
  • Sort — click column headers

Task Detail Modal

  • Description — full task description and metadata
  • Report — AI-generated markdown output rendered with tables, lists, and code blocks
  • Assets — generated source files and result files with preview and download
  • Metrics — pass/fail indicators when quantitative results are available

Status Badges

stagedqueuedsynthesizingneeds_reviewreadyexecutingfailedcompletedcancelled

File Management

Each project has its own file store backed by Google Cloud Storage. Files are accessible to the AI agent and visible in the right-panel file browser.

Upload

  • Drag-and-drop or click — in the left panel upload zone
  • Max size — 200 MB per file
  • Supported types — GDS, SDS, PDF, MD, TXT, CSV, JSON, DOC, DOCX, XLS, XLSX, PNG, JPG, SVG, PY, TS, JS, ZIP, TAR, GZ
  • MANIFEST.md — auto-generated after each upload so the AI agent knows what files exist

File Preview

  • Images — rendered inline in the browser
  • Text / Code / Markdown / CSV / JSON — full text view
  • PDF / DOCX / XLSX — text extracted and displayed
  • GDS / SDS (binary) — hex preview

AI Agent

The AI agent lives in the project chat panel. Describe what you want to test in plain language; the agent reads your design files, reasons over them, and proposes structured test tasks.

Agent Capabilities

  • File comprehension — reads and extracts content from uploaded design documents, datasheets, and spec files
  • Task awareness — queries existing tasks in the project to avoid duplicates and build on prior work
  • Test task proposals — generates structured, reviewable batches of test tasks based on your specifications
  • Task creation — commits approved tasks directly into the project pipeline

Task Proposal Flow

When the agent proposes tasks, a review card appears in chat showing each proposed task (title, description, priority). You can Approve to create all tasks, or Reject to discard and refine the prompt. Approved tasks enter the pipeline at staged status.

Streaming

Chat responses are streamed in real time so you can follow the agent's reasoning step by step.

Task Workflow

Every task moves through a defined lifecycle from creation to completion.

stagedqueuedsynthesizingneeds_reviewreadyexecutingcompleted
  • staged — task created, awaiting submission
  • queued — submitted to synthesis pipeline, waiting for a worker
  • synthesizing — AI code generation in progress
  • needs_review — awaiting engineer or technician intervention
  • ready — synthesis complete, queued for physical test execution
  • executing — running on physical hardware
  • completed — test finished successfully
  • failed — unrecoverable error; inspect logs for details
  • cancelled — stopped by an admin or technician

Phases

  • synthesis — task is in the AI code-generation pipeline
  • execution — task is in the hardware execution pipeline
  • (none) — task is staged and not yet in any pipeline

Notifications

  • Toast notifications — slide in from the top-right for success, error, and info events; auto-dismiss after 6 seconds
  • Task status watcher — polls task status in the background and fires a notification on every state transition
  • Manual dismiss — click the × on any notification to remove it immediately

Administration

The Admin panel is restricted to workspace administrators and covers member management, workspace settings, and a task overview.

Members

  • Invite — add a member by email and assign a role
  • Change role — inline dropdown per member row (Admin, User, Observer, Technician)
  • Remove — delete button removes the member from the workspace
  • Task count — shown per member for quick capacity assessment

Workspace Settings

  • Task limit — maximum active tasks per user (1–500); enforced at task creation
  • Create workspace — create a new isolated workspace; you become its admin automatically and can invite others

Task Overview Table

A read-only table of all tasks in the workspace showing task name, project, assigned user, status, and priority — useful for ops review without navigating per-project.

Roles & Permissions

RoleAccess
AdminFull access — manage members, create/edit/delete projects and tasks, view admin panel
UserCreate and edit projects and tasks; cannot manage members or access the admin panel
ObserverRead-only access to projects, tasks, and chat; no write operations
TechnicianTasks page only — sees only assigned tasks; no sidebar navigation, no project or member access

Multi-Workspace Support

A single Forge account can belong to multiple workspaces. Use the workspace switcher in the sidebar header to toggle between them — all data (projects, tasks, files, members) is fully isolated per workspace. Creating a new workspace from the Admin panel makes you its admin; you can then invite teammates by email.

Ready to get started?

Sign in to Forge and run your first agentic test in minutes.

Start Testing