Skip to content
| Marketplace
Sign in
Visual Studio Code>Machine Learning>Forge - Multi-Model Coding AgentNew to Visual Studio Code? Get it now.
Forge - Multi-Model Coding Agent

Forge - Multi-Model Coding Agent

SoftUp Agency

|
1 install
| (1) | Free
A multi-model VS Code coding agent that triages requests, runs sub-agents in parallel, and keeps spend under a hard budget cap.
Installation
Launch VS Code Quick Open (Ctrl+P), paste the following command, and press enter.
Copied to clipboard
More Info

Forge - Multi-Model Coding Agent for VS Code

Forge is a VS Code extension for coding with large language models without sending every task to the most expensive model. It triages requests, routes them to the right-sized model tier, runs sub-agents in parallel on decomposable work, applies edits deterministically, and keeps spend visible against a hard budget cap.

What it does

  • Routes each request through a cheap triage step before escalating.
  • Runs parallel sub-agents when a task can be split safely.
  • Applies file edits deterministically, with a narrow repair fallback only when a patch cannot be matched cleanly.
  • Tracks spend in real time and stops new calls when the configured cap is reached.
  • Supports attached images through a dedicated vision path.
  • Loads project memory and a persistent codebase index to keep context focused.
  • Integrates with MCP servers, hooks, and session history.

Install

npm install
npm run build

Then press F5 in VS Code with this folder open to launch an Extension Development Host with Forge loaded.

First run

  1. Run Forge: Sign In or Forge: Create Account if your workspace uses the hosted backend.
  2. Open a folder or workspace. Forge is workspace-rooted by design.
  3. Run Forge: Open in Sidebar.

Key settings

Setting Default Purpose
forge.backendUrl http://softup.agency:8282 Hosted Forge backend for auth and model proxying
forge.baseUrl empty Optional direct OpenAI-compatible provider URL
forge.budgetCapUsd 20 Hard spend ceiling
forge.warnThresholds [0.5, 0.8, 0.95] Budget warning thresholds
forge.maxConcurrentSubAgents 3 Parallel sub-agent cap
forge.maxSubAgentsPerTask 5 Max sub-agents per decomposed task
forge.maxAgentIterations 12 Tool-loop cap per agent
forge.maxEscalationsPerTask 1 Tier escalation limit
forge.projectMemoryEnabled true Load FORGE.md and .forge/rules/*.md
forge.codebaseIndexEnabled true Maintain the persistent codebase index
forge.editApprovalEnabled true Ask before applying edits

Privacy and data handling

Forge can send prompts, selected file context, tool output, and attached images to the provider or backend you configure. API keys are stored in VS Code secret storage. Workspace sessions and the codebase index are stored locally in the workspace under .forge/.

If you are preparing a marketplace submission, review the public policy draft in public/privacy.md.

Publishing assets

  • Marketplace copy
  • Privacy policy draft
  • Public asset notes

Support

  • Issues: https://github.com/myarsk/Forge-Agent/issues
  • Source: https://github.com/myarsk/Forge-Agent
  • Contact us
  • Jobs
  • Privacy
  • Manage cookies
  • Terms of use
  • Trademarks
© 2026 Microsoft