Documentation
Everything you need to captain your AI workflow and stop regressions before they board your codebase.
Quick Start
Install
npm install -g archondev Run & Start Talking
First run is conversational-first: Archon inspects the folder, loads governance, checks whether BYOK keys exist locally, and drops you into chat without a required platform login.
cd your-project && archon Plan and Execute
archon plan "Add user authentication" Plans now require design approval before execution. Content-only requests (stories, lessons, outlines, visuals) use lightweight planning so momentum stays high.
v3.0.1: Local-first chat startup, BYOK-only AI path, local usage/privacy, filesystem skills, continuity handoff, and safer parallel local execution.
Smart Onboarding: Three Scenarios
After startup, ArchonDev reads the folder and routes you to the right workflow without guesswork.
New Project
Empty folder -> guided interview that generates `.archon/active/architecture.md` and sets your governance baseline.
Adapt Existing
Existing codebase -> review-first onboarding with governance adapted to your live structure.
Continue Session
Returning session -> restores local context, surfaces last activity, and resumes pending atoms cleanly.
User Journey Governance (Lite + CLI)
ArchonDev is optimized as a full journey system, not just command docs. Both Lite and CLI flows are designed to reduce decision fatigue, prevent dead ends, and move users from setup to confident execution quickly.
Journey Sequence
- Anchor trigger, job, and desired outcome.
- Recommend the best path (continue, review, setup).
- Confirm intent and risks before implementation.
- Recover gracefully from ambiguity or failure.
- Close with a clear outcome and next action.
Why It Converts Better
- • Faster first success moment for new users
- • Clearer progress narrative in every session
- • Better trust through transparent recovery behavior
- • Lower cognitive load through explicit recommendations
CLI BYOK Journey Highlights
- •
archon modeandarchon config aikeep setup local and simple. - •
archon config airuns guided key setup with optional low-cost Gemini Flash helper prompts. - • All AI features run only with user-provided provider keys.
- • Model routing and preferences are configurable in settings for planning/reasoning/execution tiers.
- • If no key is configured, CLI provides direct recovery to BYOK setup commands.
- • Major workflow checkpoints end with clear next-best-action suggestions.
Quality Level / Posture
Set how strict your AI should be. Run lean in prototypes and lock discipline in production and enterprise.
| Posture | When to Use | AI Behavior |
|---|---|---|
| prototype | Learning, experimenting | Fast iteration, minimal governance |
| production | Real users (default) | Secure defaults, basic monitoring |
| enterprise | Mission-critical, regulated | Full governance: audit logging, RBAC, SLOs |
5-Phase Interview System
A conversational interview that guides you through defining your project, then generates a formal specification.
Discovery
Vision & users
Features
MVP scope
Technical
Tech stack
Branding
Name & style
Review
Validate & freeze
archon interview | Start or resume interview |
archon interview show | Display current Constitution |
archon interview validate | Check for completeness |
archon interview export [file] | Export to JSON |
✨ Challenge Mode
During review, the AI challenges scope creep: too many features, high complexity, vague requirements. Aims for ~20% feature reduction to keep MVPs lean.
Project Constitution
The Constitution is a formal, immutable specification generated from the interview. It's the contract between you and the AI for what gets built.
What's Inside
- • Discovery: Elevator pitch, target user, core problem
- • MVP Features: Features with user stories & acceptance criteria
- • Tech Stack: Framework, database, hosting, auth
- • Branding: Project name, tagline, colors
- • Complexity: Tier rating, estimated hours, costs
Complexity Tiers
- • SIMPLE — ≤10 routes, ≤5 tables
- • MODERATE — ≤25 routes, ≤10 tables
- • COMPLEX — ≤50 routes, ≤20 tables
- • ENTERPRISE — 50+ routes, 20+ tables
Generate prd.json from Constitution
Once frozen, transform your Constitution into executable atoms:
archon generate | Generate prd.json from frozen Constitution |
archon generate --include-post-mvp | Include POST_MVP features |
archon generate --dry-run | Preview without writing |
archon generate --output path.json | Custom output path |
CLI Reference
Use these commands to keep execution on course: approve design, execute with gates, and review in the right order.
| Command | Description |
|---|---|
archon | Interactive mode |
archon init | Initialize in your project |
archon interview | 5-phase project interview |
archon generate | Generate prd.json from Constitution |
archon plan <description> | Create a work item with AI planning |
archon execute <atom-id> | Execute with quality gates |
archon list | List all work items |
archon show <atom-id> | Show details |
archon bug report <title> | Bug report with RCA |
archon review init | Initialize code review |
archon config ai | Guided BYOK setup and key rotation flow |
archon usage | Local usage by period and model |
archon keys add <provider> | Add API key (BYOK) |
archon preferences | Settings menu |
archon mode | Choose local governance mode or BYOK AI mode |
archon index search <query> | Local semantic code search |
.archon/active/architecture.md Format
Your source of truth for governance. Uses YAML frontmatter.
---
name: My Project
version: "1.0"
boundaries:
- path: "src/db/**"
protection: STRICT
description: "Database layer"
invariants:
- id: INV-001
type: NO_DIRECT_IMPORT
path: "@supabase/supabase-js"
enforcement: BLOCKING
protectedPaths:
- pattern: "src/db/schema.ts"
reason: "Schema changes require migration"
--- DEPENDENCIES.md
Track file-level dependencies to prevent regressions.
archon deps list | View dependency rules |
archon deps add | Add a new rule |
archon deps check --files <list> | Check downstream impacts |
archon deps graph | Generate Mermaid diagram |
Code Review
AI-powered code review for any codebase. Non-destructive analysis.
archon review initCreate SQLite databasearchon review analyzeScan projectarchon review run --allAI reviews all filesarchon review planGenerate fix planarchon review exportExport reportTask Extraction Protocol
AI confirms all items from multi-part requests before starting. Nothing gets forgotten.
When you say "fix the login bug, add dark mode, and update the docs", the AI will list all 3 tasks and confirm before proceeding.
Context-Aware Intelligence
Instead of requiring slash commands, ArchonDev detects what you're doing and proactively offers help. The AI reads context — file types, error output, milestones, scope changes — and presents relevant options with a skip choice.
| Context Detected | What You Get |
|---|---|
| Bug / error / stack trace | Systematic debugging protocol |
| Frontend file changes | Design review + AI slop check |
| Feature complete / pre-commit | Fix-first review + ship readiness |
| New project / empty directory | Strategy questions before coding |
| Scope change mid-feature | Scope review (4 modes) |
| Pre-deploy / publish / launch | A11y + SEO/GEO + ship readiness |
| Milestone reached | Progress reflection |
| Complex multi-file change | Self-regulation check-in |
Every trigger includes a skip option. The AI never forces a capability — it just makes sure you know what's available.
Fix-First Code Review
Instead of listing issues for you to fix, ArchonDev auto-fixes mechanical problems and only asks about genuinely ambiguous ones.
AUTO-FIX: Unused imports, formatting issues, missing semicolons, simple type errors, console.log in production, empty catch blocks.
ASK FIRST: Ambiguous naming, logic that might be intentional, performance trade-offs, API contract changes.
BLOCK: Security vulnerabilities, architecture violations, data loss risks, broken accessibility.
Output: [N] auto-fixed | [N] need your input | [N] blockers
AI Slop Detection
AI-generated designs have recognizable patterns. ArchonDev checks frontend code against a 10-item blacklist and scores A through F.
| # | Anti-Pattern |
|---|---|
| 1 | Purple/blue gradient hero sections |
| 2 | 3-column icon grid layouts |
| 3 | Uniform border radius on everything |
| 4 | Generic stock illustrations |
| 5 | Emoji used as visual design elements |
| 6 | Drop shadow on every element |
| 7 | All text center-aligned |
| 8 | Placeholder-quality copy |
| 9 | Testimonial cards with stock avatars |
| 10 | Cookie-cutter pricing tables |
Grading: A (no patterns, intentional design) through F (7+ patterns, obvious AI slop). Score C or below triggers a flag.
Systematic Debugging
When a bug is detected, the AI follows a structured protocol instead of guessing at fixes.
Phase 1 — Root Cause Investigation: Reproduce, collect evidence, trace data flow, state the root cause. No fixes during this phase.
Phase 2 — Pattern Analysis: Check against 6 common bug patterns (state mutation, race conditions, off-by-one, type mismatch, stale reference, environment delta).
Phase 3 — Hypothesis Testing: Form hypothesis, predict outcome, apply minimal fix, verify. 3-strike rule: after 3 failed attempts, stop and ask for help.
Phase 4 — Regression Prevention: Write a regression test that fails without the fix. Check for sibling bugs elsewhere in the codebase.
Ship Readiness Dashboard
Before deploying, see a dashboard of what's been checked and what's missing.
Ship Readiness Dashboard ======================== [x] Architecture review — no violations [x] Dependency check — verified safe [x] Code review — 3 auto-fixed, 0 blockers [x] Tests — all passing [ ] Design review — NOT RUN [x] Accessibility — WCAG 2.2 AA compliant Ready to ship: NO — Design review pending
Required checks: architecture, dependencies, code review, tests. Conditional: design review (if frontend), accessibility (if web), SEO/GEO (if public), security (if API/auth).
Scope Management
When scope changes mid-feature, the AI detects it and offers four options:
| EXPAND | Add all new requirements to the current task |
| SELECTIVE | Cherry-pick: add some items, defer others |
| HOLD | Keep original scope, log new items for next task |
| REDUCE | Cut to minimum viable — ship the smallest useful version |
The AI recommends based on your project posture: prototype favors EXPAND or REDUCE; production favors HOLD or SELECTIVE.
Self-Regulation Guardrails
Hard limits that prevent the AI from going off the rails during complex tasks.
| Threshold | Action |
|---|---|
| >5 files modified | Pause — present summary and get approval |
| >15 consecutive changes | Stop — checkpoint summary, wait for go-ahead |
| >3 failed fix attempts | Stop — present what was learned, ask for guidance |
| >30 lines in one fix | Pause — explain why the fix is this large |
| Unplanned file change | Announce — explain why scope expanded |
Accessibility Check
If Archon detects a web project, it will prompt you to run A11y/SEO/GEO checks and remember your preference in .archon/config.yaml.
WCAG 2.2 AA compliance checks with legal liability warnings.
archon a11y check | Run accessibility audit |
archon a11y fix | Auto-fix common issues |
archon a11y badge | Add compliance badge |
archon a11y pre-deploy | Pre-deployment check |
SEO Optimization
archon seo check | Run SEO meta tag audit |
archon seo fix | Apply recommended fixes |
GEO for AI Search
Generative Engine Optimization — get your product cited by AI assistants.
archon geo identity | Generate brand identity phrases |
archon geo schema | Generate JSON-LD schemas |
Retired Remote Paths
Remote account, GitHub, cloud execution, saved-session, Stripe credits, and Supabase usage paths are retired. Hidden compatibility commands print a retirement message instead of calling Archon-managed Supabase or Fly services.
archon login | Retired platform auth path |
archon github connect | Retired GitHub integration path |
archon execute ATOM-001 --cloud | Retired cloud execution path |
archon session save [name] | Retired saved-session path |
Current recommendation:
- Local governance mode: governance/workflow only, no built-in AI calls
- BYOK mode: local execution with your provider keys
- Remote paths: retired for current Free/BYOK usage
Semantic Indexing
Semantic search across your codebase. The primary path is now local-first.
Local Indexing
# Install Ollama first: https://ollama.ai
ollama pull nomic-embed-text
archon index init
archon index update
archon index search "authentication flow" Retired Remote Indexing
Remote indexing through Archon-managed infrastructure is retired. Use the local index commands below; cloud index flags print retirement guidance instead of calling Supabase or Fly.
archon index init | Initialize local index |
archon index update | Index changed files locally |
archon index search <query> | Semantic search |
archon index status | Show index stats |
archon index clear | Clear index |
Smart Model Routing
Automatically routes operations to the optimal AI model tier for cost optimization. Simple tasks use cheap/fast models; complex planning uses powerful models.
Model Tiers
| Tier | Operations | Default Models | Cost |
|---|---|---|---|
| PLANNING | plan, init, design, architect_debate, interview, constitution | Claude Opus, GPT-5.2, Gemini Pro | $$$ |
| REASONING | review, validate, sentinel_check, analyze, geo, seo | Claude Sonnet, GPT-5, Gemini Pro | $$ |
| EXECUTION | execute, execute_atom, deps, a11y, format, cleanup | Claude Haiku, GPT-4o Mini, Gemini Flash | $ |
Configure Preferences
archon preferences | Open preferences menu, select "Smart Routing" |
--model <model-id> | Override routing for a single command |
💡 Cost Savings
Routing simple code execution to Haiku/Flash instead of Opus/GPT-5 can reduce costs by 80-95% with no quality loss for routine tasks.
Parallel Execution
Run multiple agents simultaneously using git worktrees with dependency-aware scheduling.
Dependency Scheduling
ArchonDev analyzes atom dependencies and file conflicts to schedule execution in parallel waves.
archon parallel schedule | Show execution schedule |
archon parallel schedule --mermaid | Output dependency graph as Mermaid |
archon parallel schedule --only-ready | Only READY atoms |
archon parallel run-waves | Execute in dependency order |
archon parallel run-waves --dry-run | Preview without executing |
Manual Parallel
archon execute --parallel ATOM-001 ATOM-002 | Execute in parallel |
archon parallel status | Show status |
archon parallel merge [atomId] | Merge worktrees |
archon parallel clean | Remove all worktrees |
Parallel Cloud (Disabled)
archon parallel cloud ATOM-001 ATOM-002 | Legacy command path (disabled in free/BYOK mode) |
History & Revert
Each atom execution creates a git commit. Revert individual atoms without affecting other changes.
CLI Users
archon history | Show atom execution history |
archon revert <atom-id> | Revert a specific atom |
archon revert <commit-hash> | Revert by git commit hash |
archon revertable | List atoms that can be reverted |
Lite Package Users
Say show atom history or revert atom to your AI. Uses git under the hood:
# List atom commits
git log --oneline | grep -E '\[(ATOM|US)-[0-9]+\]'
# Revert specific commit
git revert <commit-hash> Eject
Remove all ArchonDev metadata, leaving a clean standalone codebase you fully own.
CLI Users
archon eject | Remove ArchonDev from project |
archon eject --dry-run | Preview what would be removed |
archon eject --keep-architecture | Keep .archon/active/architecture.md |
Lite Package Users
Say eject archondev to your AI, or manually remove:
- •
.archon/,archondev-skills/,AGENTS.md - • IDE rules:
.cursorrules/.windsurfrules/CLAUDE.md/ etc.
Keep: .archon/active/architecture.md (useful standalone), progress.txt (history)
Deploy
One-click deployment with automatic platform detection.
Supported: Fly.io, Vercel, Netlify, Railway, Render
archon deploy | Auto-detect and deploy |
archon deploy --platform fly | Deploy to specific platform |
archon deploy --preview | Deploy to staging |
archon deploy --dry-run | Show what would happen |
Cleanup
Keep your workspace lean for optimal AI performance. Large files consume context.
archon cleanup check | Analyze workspace for bloat |
archon cleanup run | Execute cleanup (archive old entries, clear cache) |
archon cleanup auto enable | Enable auto check on start |
Thresholds:
- •
progress.txt> 100KB — archive old entries to docs/archive/ - •
.archon/> 10MB — clear stale cache files - • Cache files > 7 days — auto-deleted on cleanup run
Working with Existing Projects
Add governance to projects created by other AI agents or manual coding.
# Step 1: Review existing code
archon review init
archon review analyze
archon review run --all
# Step 2: Set up governance
archon init --analyze
# Step 3: Fix issues
archon plan "fix critical issues from review" VS Code Extension
Real-time architectural governance with quick-fix suggestions.
- • Real-time violation detection in Problems panel
- • Quick-fix suggestions (Ctrl+. / Cmd+.)
- • Protected path warnings
- • Boundary enforcement for cross-module imports
Installation: Download archondev-0.2.0.vsix from the downloads page, then Extensions → ... → Install from VSIX
Modes & BYOK
FREE
$0
- • Download/governance workflow
- • No built-in AI model calls
- • Works alongside your existing IDE tools
BYOK
$0 to ArchonDev
- • Use your own API keys
- • All models available
- • Archon never bills token usage
BYOK Key Security
- • Your API keys stay on your machine and are never uploaded to ArchonDev servers.
- • Keys are encrypted at rest with AES-256-GCM in
~/.archon/keys.json. - • The key file is set to owner-only permissions (0600).
- • BYOK runs locally with your own provider keys.
If your device is compromised, an attacker could access local files. Treat API keys as sensitive secrets.
Troubleshooting
"Command not found: archon"
Ensure npm global bin is in your PATH. Run npm bin -g to check.
"No API key configured"
Run archon config ai or archon keys add <provider> to enable BYOK execution.
".archon/active/architecture.md not found"
Run archon init in your project root.
Cloud execution not starting
Cloud execution is currently disabled in free/BYOK mode. Use local execution with archon execute <atom-id>.
Need Help?
Email us at techsupport@JumpingAhead.com