Package Exports
- @radekzm/bestai
- @radekzm/bestai/setup.sh
This package does not declare an exports field, so the exports above have been automatically detected and optimized by JSPM instead. If any package subpath is missing, it is recommended to post an issue to the original package (@radekzm/bestai) to support the "exports" field. If that is not possible, create a JSPM override to customize the exports field for this package.
Readme
bestAI: Omni-Vendor Convergence
The industry-leading orchestration layer for autonomous AI engineering swarms.
Strategic Value • The Toolbelt • Syndicate Model • Architecture • Quick Start
💎 The Strategic Value of bestAI (ROI)
Why do enterprises and top-tier developers choose bestAI over generic AI agents?
| Metric | Before bestAI | With bestAI v7.6 | Impact |
|---|---|---|---|
| Agent Compliance | ~6% (Advisory only) | 100% (Deterministic) | Eliminates manual review for trivial rule breaks. |
| Token Efficiency | High bloat (uncut) | -70% overhead | Drastically reduced API costs through Context OS. |
| Project Knowledge | Limited by 200k context | Infinite (2M+ RAG) | Agents can map codebases of any size (monorepo). |
| Vendor Lock-in | Single model provider | Omni-Vendor | Switch models (Claude/Gemini/Codex) based on cost/perf. |
| Consistency | "Goal Amnesia" | Zero Loss (GPS State) | Multi-session stability via Global Project State. |
🛠️ The Syndicate Toolbelt
bestAI ships with a production-ready suite of specialized tools to manage your AI workforce.
| Tool | Capability | Business Benefit |
|---|---|---|
bestai doctor |
Structural & architectural audit. | Prevents broken project states before agents start. |
bestai compliance |
Real-time agent obedience metrics. | Quantifiable audit logs of agent performance. |
bestai swarm |
Heterogeneous task dispatcher. | Parallelizes development across different AI vendors. |
bestai test |
Isolated hook testing framework. | Guarantees your security "Force Field" never breaks. |
budget-monitor |
Real-time FinOps token tracking. | Stops agents before they burn through API credits. |
session-replay |
Visual step-by-step reasoning logs. | Speeds up debugging of complex agent hallucinations. |
hook-lint |
Dependency & Latency profiling. | Ensures high-speed CLI response (Target: <200ms). |
🤖 Syndicate Decision & Collaboration Loop
[!IMPORTANT] Deterministic Force-Field: Unlike standard prompts, bestAI hooks are Fail-Closed. If an agent attempts an illegal operation, the session is physically blocked at the shell level.
🤖 The Syndicate Model (Agent Roles)
🏗️ The 5-Tier Context OS (Infinite Scalability)
We solve the "Context Overload" problem by segmenting information into distinct tiers, ensuring the agent always knows the most critical state without drowning in data.
- T0 (HOT): The Global Project State (GPS.json). The project's brain. Shared by all swarm agents.
- T1 (WARM): T3-Summary.md. The map of the entire codebase. Hierarchical and dense.
- T2 (COOL): The active module files the agent is editing.
- T3 (COLD): The rest of the codebase, accessed via RAG-native router.
- T4 (FROZEN): Configs and secrets protected by Deterministic Hooks.
⚡ Quick Start
1. Global Installation
npm install -g @radekzm/bestai
# or use npx instantly:
npx @radekzm/bestai@latest init .2. Dispatch Tasks to the Syndicate
# Research task for Gemini (utilizing 2M context)
bestai swarm --task "Find all deprecated auth calls" --vendor gemini
# Coding task for Claude
bestai swarm --task "Replace auth calls using results in GPS.json" --vendor claude🔄 Cross-Tool Compatibility
Already using Cursor or Windsurf? Translate bestAI standards instantly:
bestai generate-rules --format cursor
Built for the next generation of autonomous engineering.
License: MIT | radekzm & the bestAI Swarm