Forge
- Engineering reasoning
- Independent review
- Second opinion
- Durable engineering context
Preview
Forge works inside your IDE, beside your coding agent. It independently investigates the same codebase, challenges plans and fixes, preserves engineering context, and gives your agent better instructions when implementation is needed.
Forge is not another coding agent. It reasons. The agent implements.
Forge 0.2.1 Preview
Free. Bring your own OpenAI or Anthropic API key, and choose the provider and model in Forge.
How it works
The coding agent proposes or implements. Forge independently investigates and challenges. The codebase is the shared evidence.
Forge
Your code
Coding agent
Forge and the coding agent both look at your code. They do not stand in for each other.
Workflow
A typical pass looks like this. The details change; the roles do not.
The agent works from the request in front of it. That plan can look complete and still miss a repository fact.
Forge inspects the workspace on its own. It does not inherit the agent’s framing.
Architectural, runtime, deployment, or context problems become visible while they are still cheap to correct.
You get a reasoned account of what holds, what does not, and why — with the codebase as evidence.
Forge can hand the coding agent a tighter brief so the next change matches the repository you actually have.
Current capabilities
These are the capabilities in the current Preview.
Ask engineering questions about the active repository. Forge investigates the workspace and answers from the code, not from the agent’s last guess.
Independently review a coding agent’s plan or fix. Forge challenges assumptions before they harden into the wrong implementation.
Forge reads the same codebase as your agent, then forms its own picture of architecture, runtime behavior, and deployment reality.
Keep conversations in an engineering knowledge workspace on this machine. A Forge Project is not a repository, an editor workspace, or a folder. Threads that are not in a project yet stay available.
Preserve durable engineering findings in the workspace so the next review does not start from zero. That memory is separate from Forge Projects and conversations.
When implementation is needed, Forge writes clearer instructions for the coding agent — grounded in what the repository actually does.
Project Memory
Forge can keep durable findings with the project so later reviews start from known constraints instead of rediscovering them.
Project memory is stored locally under .forge/ in the
workspace. It is separate from Forge Projects and conversations,
which stay in the Forge knowledge directory on this machine. There
is no automatic cloud or team sync.
Privacy and AI access
Forge uses an API key you provide for OpenAI or Anthropic. You choose the provider and model in Forge.
You provide the key for the provider you select. Forge stores OpenAI and Anthropic keys separately in VS Code SecretStorage. When Forge reasons, it sends the context required for that request directly to the selected provider.
Project memory lives under .forge/. Conversations,
folders, tags, and attachments live in the Forge knowledge
directory on this machine, outside editor extension storage, and
do not sync to the cloud. Usage tracking is local. Forge does not
send telemetry to DevFoundry. Repository tools do not read
contents of .env* files.
“Local-first” means the product works from your machine and your workspace. It does not mean the AI model runs on your computer.
Access
Forge 0.2.1 is available in Preview. You bring your own OpenAI or Anthropic API key and choose the provider and model in Forge.
Preview
The current release. Forge runs as a VS Code extension in VS Code and compatible editors such as Cursor. Bring your own OpenAI or Anthropic API key.
Download Forge 0.2.1 Visual Studio Marketplace Open VSX / Cursor Installation & setup
Forge 0.2.1 is on the Visual Studio Marketplace, the Open VSX Registry, and is also distributed directly as a VSIX from this site. Gallery listings are not required to use Forge.