NeuraForge — AI-Driven Engineering

🇬🇧 English | 🇫🇷 Français

NeuraForge orchestrates specialized AI agents to turn a request into delivered, tested, and traced code. You describe what you want, NeuraForge distributes the work across 6 agents (Analyst, Planner, Dev, QA, DocOps) and delivers code with full traceability of AI decisions.

Built for enterprise development teams: provider-agnostic (Claude Code, Gemini, Copilot, OpenCode, custom API), injectable skills and rules, security filtering, lessons learned (retex), built-in compliance, feature-first for brownfield projects.

Quickstart (5 minutes)

Prerequisites

• Node.js 20+ and bun (or npm/yarn)
• At least one AI provider installed: Claude Code, Gemini CLI or GitHub Copilot CLI

1. Install

$ bun add -g @neuravim/neuraforge
$ neuraforge --version    # 0.10.0

2. Initialize in your project

$ cd my-project
$ neuraforge init

NeuraForge scans the project and automatically detects the tech stack, architecture, CI/CD and package manager. Everything is stored in .aiden/ (commit it with your code).

3. First workflow (flash = bug fix / small feature)

$ neuraforge flash "Add a GET /health endpoint returning {status: ok}"

Flash mode is direct: Dev implements, writes tests, traces the decision. No ceremony.

4. Full workflow (standard = feature with analysis + review)

$ neuraforge standard "Add a user management REST API with pagination"

Standard mode chains 4 automatic phases:

| Phase | Agent | What it does | |-------|-------|-------------| | Discovery | Analyst | Scans code, estimates complexity, identifies risks | | Planning | Planner | Creates a spec + stories with acceptance criteria | | Implementation | Dev | Implements each story with tests | | Validation | QA | Reviews code vs criteria, loops with Dev on failure |

Artifacts generated in .aiden/features/<slug>/: brief.md, spec.md, stories/, review.md, change-summary.md.

5. Verify

$ neuraforge status                        # View all features
$ neuraforge metrics                       # Tokens, costs, durations
$ cat .aiden/memory/decisions.jsonl   # AI decision history

3 adaptive modes

| Mode | Usage | Agents | Estimated time | |------|-------|--------|---------------| | neuraforge flash | Bug fix, hotfix, small feature | Lead + Dev | < 30 min | | neuraforge standard | Medium feature, new API | + Analyst, Planner, QA | 2h - 1 day | | neuraforge enterprise | Critical project, compliance required | + DocOps + human approval | 1 day+ |

NeuraForge automatically detects the optimal mode. You can switch modes mid-workflow without losing work:

$ aiden switch export-csv standard    # switch from flash to standard mid-feature

Intelligent multi-provider

Each agent receives the optimal provider for its task:

| Task | Default provider | Reason | |------|-----------------|--------| | Analysis & Review | Claude Code | Best reasoning | | Code generation | Gemini | Performant multi-file | | Quick fix | Copilot | Speed + fixed cost |

4 routing strategies: fixed, prefer, auto, cost-optimized. Automatic fallback if a provider is unavailable.

For teams

Enterprise rules

Define your standards in .aiden/rules/ (coding standards, security, architecture). They are automatically injected into every agent's prompts.

$ neuraforge rules add security-policy.md

Skills (extensibility)

Skills are declarative packages (YAML + Markdown) that bundle instructions, rules and hooks. Drop a folder in .aiden/skills/ and NeuraForge discovers it automatically.

$ neuraforge skills init my-skill           # Create a skill
$ neuraforge skills list                    # List skills
$ neuraforge skills enable/disable <name>   # Enable/disable

3 built-in templates: code-review-strict, compliance-rgpd, jira-sync.

Multi-level configuration

5 configuration levels that cascade:

NeuraForge defaults → Global config (~/.aiden/) → Project config (.aiden/) → Feature config → CLI flags

Global config is installed once per workstation and applies to all projects.

Compliance and audit

• decisions.jsonl: every AI decision is traced (who, what, why, token count)
• metrics.jsonl: cost per agent, per provider, per feature
• Human approval in enterprise mode between each phase
• DocOps generates changelog, documentation, runbooks

Commands

neuraforge init                          # Initialize NeuraForge in the project
neuraforge flash "description"           # Fast mode (direct Dev)
neuraforge standard "description"        # Standard mode (Analyst → Planner → Dev → QA)
neuraforge enterprise "description"      # Enterprise mode (+ DocOps + approval)
neuraforge status [feature]              # Feature status
neuraforge resume <feature>              # Resume a blocked feature
neuraforge abort <feature>               # Abort a feature
aiden switch <feature> <mode>       # Switch mode mid-feature
neuraforge agent <name> "prompt"         # Invoke an agent directly
neuraforge provider status|switch|cost   # Manage AI providers
neuraforge metrics [--export csv]        # Usage metrics and costs
neuraforge config show|set               # Configuration
neuraforge rules add|list|remove         # Enterprise rules
neuraforge skills list|init|add|remove   # Extensibility skills
neuraforge retex add|list|search|remove  # Lessons learned
neuraforge template customize|list       # Artifact templates

Claude Code mode (slash commands)

NeuraForge also works directly inside Claude Code without installing the CLI:

/aiden-init        /aiden-flash       /aiden-standard    /aiden-enterprise
/aiden-status      /aiden-resume      /aiden-agent       /aiden-rules

Both modes (CLI and slash) write to the same .aiden/ files.

Documentation

| Guide | Description | |-------|-------------| | Installation | Prerequisites, installation, provider setup | | Quickstart | Complete scenario from zero in 5 minutes | | First feature | Detailed standard workflow phase by phase | | Deployment guide | Deploy NeuraForge in an organization (pilot, rollout, governance) | | Modes | Flash, standard, enterprise: when to use what | | Rules & templates | Configure enterprise standards | | Skills | Enterprise extensibility: instruction packages, rules and hooks | | Brownfield | Add NeuraForge to an existing project | | Team usage | Git workflow, onboarding, conventions | | CI/CD | Pipeline integration |

Note: The documentation site is written in French (project language for docs). The code, CLI, and this README are in English.

Tests

$ bun test            # 444 unit tests
$ bun run typecheck   # TypeScript strict check
$ bun run lint        # ESLint

License

AGPL-3.0 — See LICENSE for details.