raceos-factory 1.0.5

AI Software Factory for the RaceOS motorsport platform

RaceOS AI Software Factory

AI-native software factory for the RaceOS motorsport platform.

---

How It Works

┌─────────────────────────────────────────────────────────────────────┐
│                                                                     │
│   Human: "implement engine load gauge"                              │
│     │                                                               │
│     ▼                                                               │
│   ┌──────────┐    ┌──────────┐    ┌──────────┐    ┌──────────┐    │
│   │   CLI    │───►│ 9Router  │───►│ LangGraph│───►│ OpenCode │    │
│   │ (raceos) │    │(select   │    │(orchestr)│    │(execute) │    │
│   └──────────┘    │ model)   │    └──────────┘    └──────────┘    │
│                   └──────────┘         │                │          │
│                        │               ▼                ▼          │
│                        │         ┌──────────┐    ┌──────────┐     │
│                        │         │  Brain   │    │LLM Gateway│     │
│                        │         │(.raceos/)│    │(9Router│     │
│                        │         └──────────┘    │ / 9Router)│     │
│                        │                         └──────────┘     │
│                        │                              │            │
│                        ▼                              ▼            │
│                   ┌──────────────────────────────────────┐        │
│                   │  LLM (Claude / GPT-5 / Gemini / DS)  │        │
│                   └──────────────────────────────────────┘        │
│                                                                     │
└─────────────────────────────────────────────────────────────────────┘

---

Complete Flow

Human
  │
  │  $ raceos firmware "fix load always showing 100%"
  │
  ▼
┌─ CLI ──────────────────────────────────────────────────────────────┐
│  1. Parse input → intent: "debug", domain: "firmware"              │
│  2. Create task → TASK-A1B2C3                                      │
│  3. Save checkpoint (crash recovery)                               │
└────────────────────────────────────────────────────────────────────┘
  │
  ▼
┌─ 9Router (model selection) ────────────────────────────────────────┐
│  4. Task type: debugging → select Claude Sonnet 4.6, temp=0        │
│  5. Check budget: $3.42/$50.00 (OK, no downgrade)                  │
│  6. Set timeout: 120s                                              │
└────────────────────────────────────────────────────────────────────┘
  │
  ▼
┌─ LangGraph (orchestration) ────────────────────────────────────────┐
│                                                                    │
│  7. ORCHESTRATOR → classify: type=debugging, complexity=medium     │
│  │                                                                 │
│  8. CONTEXT AGENT → load tiered context:                           │
│  │   Tier 1: .raceos/ principles (cached)              2K tokens  │
│  │   Tier 2: openspec/specs/display-dashboard/         3K tokens  │
│  │   Tier 3: firmware/middleware/telemetry_parser.cpp   4K tokens  │
│  │   Tier 4: codebase memory patterns                  1K tokens  │
│  │                                                                 │
│  9. ENGINEER → dispatch to OpenCode                                │
│  │                                                                 │
│  10. REVIEWER → verify code quality (Claude, temp=0)               │
│  │                                                                 │
│  11. VALIDATOR → trust executor (tests already passed)             │
│  │                                                                 │
│  12. KNOWLEDGE → store patterns + suggest follow-ups               │
│                                                                    │
└────────────────────────────────────────────────────────────────────┘
  │
  ▼
┌─ OpenCode (execution) ─────────────────────────────────────────────┐
│  13. Receive: context (10K tokens) + task + constraints            │
│  14. Tools: file.read → file.write → terminal.run → git.commit    │
│  15. LLM calls → via Gateway → Claude generates fix               │
│  16. Run: pio build ✅ → pio test ✅ (16 passed)                    │
│  17. Commit: [factory] fix: convert raw byte to percentage         │
└────────────────────────────────────────────────────────────────────┘
  │
  ▼
┌─ LLM Gateway (HTTP) ──────────────────────────────────────────────┐
│  18. POST /chat/completions → model: claude-sonnet-4.6             │
│  19. Retry: 429→backoff, 5xx→retry, 400→fail                      │
│  20. Fallback: Claude fails → try GPT-5 → try DeepSeek            │
│  21. Cost: track tokens, record to ledger                          │
└────────────────────────────────────────────────────────────────────┘
  │
  ▼
Human sees:
  ✅ Fixed: raw byte (0-255) not converted to percentage
  📁 Modified: telemetry_parser.cpp, board_config.h, test_telemetry_parser.cpp
  💰 Cost: $0.20
  ⏱️  Duration: 44s
  💡 Follow-up: "check if rpm, temp, duty also need conversion"

---

Components

Component	Role	Location
CLI	Human interface (15 commands + shell)	src/factory/cli/
9Router	Select model + enforce budget	src/factory/router/
LangGraph	Orchestrate 11 AI agents	src/factory/orchestrator/
OpenCode	Execute code via MCP tools	src/factory/executor/
LLM Gateway	HTTP proxy to LLM providers	src/factory/gateway/
Context Agent	Load Brain + specs + memory	src/factory/context/
OpenSpec	Parse/generate specs	src/factory/spec_engine/
Codebase Memory	Store implementation patterns	src/factory/memory/
Project Brain	Immutable knowledge (read-only)	../.raceos/

---

Install

pip install raceos-factory

Setup

# 1. Get LLM API key (choose one gateway):
#    9Router: https://9router.ai/keys
#    9Router:    https://9router.ai

# 2. Configure
cd your-raceos-project/raceos-factory
cp .env.example .env
# Edit .env → add your API key

# 3. Verify
raceos doctor

See LLM Gateway Setup Guide for detailed options.

---

Usage

Interactive Shell

$ raceos

raceos> implement engine load gauge
raceos> fix typo in ecu_parser.h
raceos> how to wire display to stm32
raceos> /status
raceos> /budget
raceos> /quit

Direct Commands

raceos feature "add BLE support" --domain firmware
raceos firmware "fix watchdog timeout"
raceos hardware "wiring diagram for racepanel"
raceos docs "explain HAL interfaces"
raceos review
raceos doctor
raceos install
raceos update

---

AI Agents (11)

Agent	Handles	Model
Orchestrator	Classify + route tasks	DeepSeek (cheap)
Context	Load Brain + specs + memory	No LLM (file I/O)
Planner	Generate specs before execution	GPT-5
Engineer	Write/fix code	Claude
Reviewer	Independent code review	Claude (temp=0)
Validator	Run build + tests	No LLM (subprocess)
Knowledge	Learn patterns + suggest follow-ups	No LLM
Architect	Design systems + ADRs	GPT-5
Research	Market/user evidence	GPT-5
Product	Problem definition + MVP	GPT-5
Hardware	Wiring/BOM/pinout	Gemini

---

Cost

Task Type	Cost	Time
Feature (complex)	$0.54	~56s
Bug fix	$0.20	~44s
Documentation	$0.13	~21s
Classification	$0.0001	~2s

Daily budget: $50 (configurable). Auto-downgrades model when approaching limit.

---

Architecture

Project Brain (.raceos/) ── read-only, immutable
         ↓
    OpenSpec (specs/) ── behavioral contracts
         ↓
    LangGraph (DAG) ── orchestrate 11 agents
         ↓
    OpenCode (tools) ── file, terminal, git, patch
         ↓
    9Router (rules) ── deterministic model selection
         ↓
    LLM Gateway ── HTTP to providers
         ↓
    LLM (Claude/GPT/Gemini/DeepSeek)

Zero circular dependencies. Single responsibility per component.

---

Configuration

File	Controls
.env	API key (secret)
config/router.yaml	Model routing + budget + fallback + retry
config/factory.yaml	Paths + workflow + logging
config/opencode.json	Execution tools + safety
config/context-tiers.yaml	Context budgets per tier
config/memory.yaml	Codebase memory indexing
config/openspec.yaml	Spec engine settings

See CONFIG_INDEX.md for "I want to change X → edit this file".

---

Documentation

Doc	Purpose
LLM Gateway Setup	Choose + configure gateway
Architecture	Complete system flow
CLI Design	CLI commands + shell
Integration	How components connect
Deployment	Production deployment
CONFIG_INDEX	Configuration reference

---

Contributing

See CONTRIBUTING.md.

Security

See SECURITY.md.

License

MIT — see LICENSE.