Forge AI

Forge better code. Faster.

Open-source multi-agent coding agent for the terminal. Built for speed, cost-efficiency, and operator control with DeepSeek models.

What Forge AI Is

Forge AI is a terminal-first coding agent with a built-in multi-agent workflow:

• Architect plans the change.
• Coder implements and runs checks.
• Reviewer validates for quality and correctness.

It also includes local semantic code search (TF-IDF), persistent project memory, checkpoint/rollback, and quality gates for safer iteration.

Capability Snapshot

| Capability | Forge AI | Cursor Composer 2.5 | Claude Code | |---|:---:|:---:|:---:| | Multi-agent pipeline (Architect → Coder → Reviewer) | ✅ | ⚠️ Varies by workflow | ❌ | | Local semantic search (TF-IDF RAG) | ✅ | ✅ | ❌ | | Project memory (local) | ✅ | ⚠️ Limited | ✅ | | Checkpoint and rollback | ✅ | ❌ | ❌ | | Terminal-native workflow | ✅ | ⚠️ Hybrid | ✅ | | Open-source core | ✅ | ❌ | ❌ | | Model routing controls | ✅ | ⚠️ Limited | ⚠️ Limited |

Notes:

• Table reflects current Forge AI features in this repository.
• Competitor columns are a practical high-level snapshot and may evolve over time.

Install

npm install -g @0x0r10n/forge-ai
# or
bun add -g @0x0r10n/forge-ai

Quickstart

# one-time setup
forge-ai setup

# interactive mode
forge-ai

# start with a task
forge-ai "add JWT authentication to my Express app"

Verify setup:

forge-ai doctor

CLI Commands

| Command | Description | |---|---| | forge-ai setup | Configure API key and defaults | | forge-ai config | View/update configuration | | forge-ai models | Show available model options | | forge-ai doctor | Run environment and dependency checks | | forge-ai update | Update global install |

Slash Commands

| Command | Description | |---|---| | /multi <task> | Run Architect → Coder → Reviewer pipeline | | /oneshot <task> | Run pipeline and exit with CI-friendly status | | /think <question> | Force deeper reasoning model | | /checkpoint [label] | Snapshot project files | | /commit [hint] | Generate smart conventional commit | | /memory [query] | Recall saved project memory | | /reindex | Rebuild semantic index | | /report [days] | Usage report | | /status | Session status | | /ship | Run production quality gate | | /cost | Quick token usage display | | /clear | Clear session history | | /help | Show command list |

Quality Gate

Forge AI includes a built-in ship gate (/ship) that checks local release readiness:

• git cleanliness (when in a git repo)
• typecheck (if script exists)
• lint (if script exists)
• tests (if script exists)
• build (if script exists)

This is designed to reduce “looks good but fails in CI” outcomes.

Configuration

Config location:

~/.forge/config.json

Example:

{
  "apiKey": "sk-...",
  "defaultModel": "deepseek-chat",
  "autoRouteModels": true,
  "autoIndexOnStartup": true,
  "requireConfirmationForDangerous": true,
  "maxIterations": 30
}

Current Scope and Limits

What is production-usable now:

• terminal workflow and slash command surface
• multi-agent orchestration
• local TF-IDF semantic search
• memory/checkpoint storage under ~/.forge
• packaging as @0x0r10n/forge-ai with forge-ai binary

Known limits:

• no MCP integration yet
• no native web retrieval tool by default
• benchmark/comparison claims are not a formal public benchmark suite

Development

git clone [email protected]:0x0r10n/forge-ai.git
cd forge-ai
bun install
bun run typecheck
bun run build
bun dev

License

MIT © Forge AI Contributors