Skip to content
FORJA v2.0[pt-br]
Follow development
FORJA 2.0 · IN DEVELOPMENT

Keep your AI work moving. Know what happens next.

An opinionated AI agent development environment built around the RPI workflow, from specification to review, with you in control of every stage.

Follow developmentThe 2.0 MVP is in development. Preview the direction below.
Design preview · planned MVP: Login with OAuth · Implementer

Login with OAuth · Implementer · Running · Send an instruction, @ to select an agent · ~/dev/projects/forja

WHY FORJA

The work spans more than a conversation.

Research produces decisions. Planning turns them into tasks. Implementation changes code. Tests and reviews send work back for correction. Keeping those steps connected across terminals and documents takes attention away from the feature.

Forja 2.0 is being built around an opinionated RPI workflow, connecting the spec, plan, agents and execution history with your explicit approval before the next stage.

THE GUIDED WORKFLOW

Five roles. Explicit handoffs.

  1. 01 · Research

    Clarify the problem and write the spec.

  2. 02 · Planner

    Turn the approved spec into a plan.

  3. 03 · Implementer

    Build from the approved plan.

  4. 04 · Tester

    Run checks and report failures.

  5. 05 · Reviewer

    Review the changes and record findings.

Corrections return to the Implementer with links to the request and its execution. Automatic handoffs remain bounded; approvals stay explicit.

Approved

PLANNED FOR THE MVP

Stay close to the work.

One tab per feature

Keep each feature’s conversations, target repository and worktree together. Work on multiple features without mixing their context.

Feature tabs: Renew session, Quick search

Execution you can follow

Watch agent commands and output through the Forja MCP runner. Use your own shell in a separate tab, and keep background processes visible in the footer.

Agent terminals: dev, tests and zsh with the collapsed footer

Documents you can discuss

Read specs, plans and reviews as formatted Markdown with Mermaid. Select a passage to prepare a revision request for its author.

SPEC-042 document with the selected passage and the Review with agent action

Your tools, configured per role

Choose the CLI, model, instructions and local skills for each of the five agents. Skills stay read-only; your existing tools keep their authentication.

Roles and CLI configuration: Implementer (Codex) and Reviewer (agy)

Target integrations: Claude Code · Codex · OpenCode · agy · Grok

Availability depends on validated adapter capabilities.

A focused first release.

The first MVP focuses on the Guided workflow and a Linux acceptance environment. Full autonomous approvals, onboarding and cross-platform release packaging are outside this initial scope.

Q: Can I download Forja 2.0 now?

The 2.0 MVP is still in development. Follow the repository for progress and release information.

Q: Does Forja replace my AI CLI?

Forja defines the RPI workflow and your approval points. Installed coding CLIs provide execution, with per-role configuration and validated controlled-execution capabilities.

Q: Who decides when work moves forward?

You approve the spec, plan and key transitions. Agent handoffs can be enabled, but do not bypass required approvals or permission to integrate changes.

Follow Forja 2.0 as it takes shape.

An opinionated AI agent development environment built around the RPI workflow, from specification to review, with you in control of every stage.

Follow development