FrootAI Architecture Overview
> System design, data flow, and component architecture of the FrootAI platform.
1. System Architecture
Component Summary
| Component | Technology | Transport | Artifact |
|---|---|---|---|
| Website | Docusaurus 3, React, TypeScript | HTTPS | Static site on GitHub Pages |
| MCP Server | Node.js, JSON-RPC 2.0 | stdio | npm package (frootai-mcp) |
| VS Code Extension | TypeScript, VS Code API | In-process | VSIX on Marketplace |
| Knowledge Hub | Markdown + Mermaid | File system | docs/*.md |
| Solution Plays | Markdown, JSON, Python, YAML | File system | solution-plays/*/ |
2. The 6 Layers
FrootAI is organized in 6 conceptual layers, from foundational infrastructure to user-facing solutions:
| Layer | Name | Contents |
|---|---|---|
| 1. Knowledge | FROOT Framework | 18 modules across 5 layers (F·R·O·O·T) |
| 2. Tooling | Developer Kit | MCP Server (25 tools) + VS Code Extension (13 commands) |
| 3. Scaffolding | Agentic OS | .github/ files — agent rules, prompts, CI, templates |
| 4. Configuration | TuneKit | config/ — models, guardrails, routing, search, chunking |
| 5. Evaluation | Quality Gates | evaluation/ — golden sets, scoring scripts, benchmarks |
| 6. Solutions | Plays | 20 pre-built scenario accelerators |
3. Data Flow
How a user’s request flows through the system:
Request Types
| Type | Flow | Tools Used |
|---|---|---|
| Knowledge Query | User → Chat → MCP → Docs → Response | search_knowledge, get_module, lookup_term |
| Agent Build | User → Chat → MCP → Plays + Docs → Scaffold | agent_build |
| Agent Review | User → Chat → MCP → Analysis → Findings | agent_review |
| Parameter Tune | User → Chat → MCP → Config → Optimized | agent_tune |
| Azure Docs | User → Chat → MCP → Azure → Response | fetch_azure_docs |
4. DevKit + TuneKit Model
FrootAI uses a two-part approach to make projects agent-ready:
| Aspect | DevKit | TuneKit |
|---|---|---|
| Purpose | Define what the agent does | Configure how well it does it |
| Location | .github/ | config/ + evaluation/ |
| Changes | Per-project, rarely | Per-iteration, frequently |
| Content | Markdown rules, YAML workflows | JSON parameters, Python scripts |
| Analogy | The recipe | The seasoning |
5. .github Agentic OS
The Agentic OS is structured around 7 primitives organized in 4 layers:
The 7 Primitives
| # | Primitive | File(s) | Purpose |
|---|---|---|---|
| 1 | Agent Rules | agent.md | Behavioral boundaries and instructions |
| 2 | Context | copilot-instructions.md | Project knowledge for AI assistants |
| 3 | Prompts | prompts/*.prompt.md | Reusable, parameterized prompt templates |
| 4 | Workflows | workflows/*.yml | CI/CD automation pipelines |
| 5 | Templates | ISSUE_TEMPLATE/, pull_request_template.md | Structured collaboration |
| 6 | Config | config/*.json | Tunable parameters |
| 7 | Evaluation | evaluation/ | Quality benchmarks and scoring |
Composition
Primitives are independent but synergistic:
agent.mdalone = basic agent behavioragent.md+copilot-instructions.md= context-aware agent- All 7 primitives = fully-equipped AI-native project
6. MCP Server Architecture
Tool Groups
| Group | Count | Network Required | Description |
|---|---|---|---|
| Static | 4 | No | Query bundled knowledge — fast, offline |
| Live | 2 | Yes | Fetch real-time external documentation |
| Chain | 3 | No | Multi-step agent workflows (build → review → tune) |
| AI Ecosystem | 5+ | No | Architecture patterns, model guidance, platform info |
Bundle
The npm package bundles all knowledge (664KB) so the server works offline. No database, no API keys, no external dependencies at runtime.
7. VS Code Extension Architecture
Key Design Decisions
| Decision | Rationale |
|---|---|
| Standalone engine | Works without MCP server or network |
| Bundled knowledge | No external fetching for core features |
| 24h cache TTL | Balance freshness vs. offline reliability |
| Layer colors | Visual identification of FROOT layers |
| Tool grouping | Logical organization matches MCP server groups |
8. Deployment Architecture
Deployment Channels
| Target | Artifact | Trigger | URL |
|---|---|---|---|
| GitHub Pages | Static site | Push to main (website/) | frootai.dev |
| npm Registry | Node.js package | Release tag (v*) | npmjs.com/package/frootai-mcp |
| VS Code Marketplace | VSIX extension | Release tag (v*) | marketplace.visualstudio.com |
| GitHub Releases | Release notes + assets | Release tag (v*) | github.com/frootai/frootai/releases |
No Backend Required
FrootAI is entirely static:
- Website = pre-built HTML/CSS/JS
- MCP Server = local stdio process
- VS Code Extension = local extension
- No databases, no cloud functions, no API servers
This zero-backend architecture means:
- Zero hosting cost (GitHub Pages is free)
- Zero latency for core operations
- Zero downtime (static files never crash)
- Zero security surface (no attack vectors)
> Next: Admin Guide · User Guide · API Reference
Last updated on