marketing-shibata50/multica
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

docs: slim down README and split into topic-specific docs

Browse source 
Add local full-stack development section (pnpm dev:local) and
move detailed content (credentials, CLI, skills/tools, time
injection, development guide) into separate docs/ files.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
Jiayuan Zhang 2026-02-15 03:23:54 +08:00
parent 058af56d47
commit 18a6996c97
6 changed files with 250 additions and 198 deletions
Download patch file Download diff file Expand all files Collapse all files

242
README.md
View file

@ -9,31 +9,27 @@ See [Memo](./docs/memo.md) for product vision, architecture, and roadmap.
## Project Structure ## Project Structure
``` ```
src/
├── agent/ # Core agent module
│ ├── context-window/ # Token-aware context management
│ ├── profile/ # Agent profile management
│ ├── session/ # Session persistence with compaction
│ ├── skills/ # Modular skill system
│ └── tools/ # Agent tools
│ └── web/ # Web fetch and search tools
├── gateway/ # WebSocket gateway for remote access
├── hub/ # Agent coordination hub
└── shared/ # Shared types
apps/ apps/
├── desktop/ # Electron desktop app (recommended) ├── cli/ # Command-line interface
└── web/ # Next.js web application ├── desktop/ # Electron desktop app (recommended)
├── gateway/ # NestJS WebSocket gateway
├── server/ # NestJS REST API server
├── web/ # Next.js web app
└── mobile/ # React Native mobile app
packages/ packages/
├── sdk/ # Gateway client SDK ├── core/ # Agent engine, hub, channels
├── store/ # Zustand state management ├── sdk/ # Gateway client SDK
└── ui/ # Shared UI components ├── ui/ # Shared UI components (Shadcn/Tailwind v4)
├── store/ # Zustand state management
├── hooks/ # React hooks
├── types/ # Shared TypeScript types
└── utils/ # Utility functions
skills/ # Bundled skills (commit, code-review) skills/ # Bundled agent skills
``` ```
## Getting Started ## Quick Start
```bash ```bash
pnpm install pnpm install
@ -42,149 +38,38 @@ pnpm install
### Development ### Development
```bash ```bash
# Desktop app (recommended for local development) pnpm dev # Desktop app (standalone, no Gateway needed)
pnpm dev pnpm dev:gateway # Gateway only
pnpm dev:web # Web app only
# Web app (for browser-based access) pnpm dev:all # Gateway + Web
pnpm dev:web # Start Web app on :3000
# Gateway (for remote/mobile clients)
pnpm dev:gateway # Start Gateway on :3000
pnpm dev:all # Start both Gateway and Web app
``` ```
The Desktop app runs a standalone Hub with embedded Agent Engine - no Gateway required for local use. ### Local Full-Stack Development
### Environment Configuration `pnpm dev:local` starts the entire stack locally (Gateway + Desktop + Web) with isolated data directories, useful for end-to-end development and testing.
**Desktop** (`apps/desktop/.env.*`): **Setup:**
| Variable | Description | 1. Copy `.env.example` to `.env` at the repo root
|----------|-------------| 2. Fill in `TELEGRAM_BOT_TOKEN` (get from [@BotFather](https://t.me/BotFather))
| `MAIN_VITE_GATEWAY_URL` | WebSocket Gateway URL for remote device pairing | 3. Run `pnpm dev:local`
| `MAIN_VITE_WEB_URL` | Web app URL for OAuth login redirect |
**Web** (`apps/web/next.config.ts`): **What it starts:**
| Variable | Description | | Service | Address | Notes |
|----------|-------------| |---------|---------|-------|
| `API_URL` | Backend API URL (default: `https://api-dev.copilothub.ai`) | | Gateway | `http://localhost:4000` | Telegram long-polling mode |
| Web | `http://localhost:3000` | OAuth login flow |
| Desktop | — | Connects to local Gateway + Web |
**Build for different environments:** **Data isolation:** All data goes to `~/.super-multica-dev` and `~/Documents/Multica-dev`, separate from production `~/.super-multica`.
**Related commands:**
```bash ```bash
# Desktop pnpm dev:local:archive # Archive dev data and start fresh
pnpm --filter @multica/desktop build # Production (.env.production)
pnpm --filter @multica/desktop build:staging # Staging (.env.staging)
# Web (Vercel)
# Set API_URL in Vercel Dashboard → Settings → Environment Variables
``` ```
See `apps/desktop/.env.example` and `apps/web/.env.example` for details.
### Monorepo Development
| Command | Purpose |
|---------|---------|
| `pnpm dev` | Full dev mode — watches `core`, `types`, `utils` packages |
| `pnpm dev:desktop` | Desktop only — skip package watching |
**When modifying packages:**
1. Edit code in `packages/core`, `packages/types`, or `packages/utils`
2. Terminal shows `[core] ESM ⚡️ Build success` (~100ms)
3. Restart Desktop to apply changes (Ctrl+C, then `pnpm dev`)
> **Why restart?** Electron main process does not support hot reload — this is an Electron limitation, not ours.
### Credentials
```bash
multica credentials init
```
Creates:
- `~/.super-multica/credentials.json5` — LLM providers + tools
- `~/.super-multica/skills.env.json5` — skill/plugin API keys
Example `credentials.json5`:
```json5
{
version: 1,
llm: {
provider: "openai",
providers: {
openai: { apiKey: "sk-xxx", model: "gpt-4o" }
}
},
tools: {
brave: { apiKey: "brv-..." }
}
}
```
### LLM Providers
**OAuth Providers** (external CLI login):
- `claude-code` — requires `claude login`
- `openai-codex` — requires `codex login`
**API Key Providers** (configure in `credentials.json5`):
- `anthropic`, `openai`, `kimi-coding`, `google`, `groq`, `mistral`, `xai`, `openrouter`
Check status: `/provider` in interactive mode
## CLI
```bash
multica # Interactive mode
multica run "prompt" # Single prompt
multica chat --profile my-agent # Use profile
multica --session abc123 # Continue session
multica session list # List sessions
multica profile list # List profiles
multica skills list # List skills
multica help # Show help
```
Short alias: `mu`
## Sessions
Sessions persist to `~/.super-multica/sessions/<id>/` with JSONL message history and JSON metadata. Context windows are automatically managed with token-aware compaction.
## Profiles
Profiles define agent identity, personality, and memory in `~/.super-multica/agent-profiles/<id>/`.
```bash
multica profile new my-agent # Create profile
multica profile list # List all
multica profile edit my-agent # Open in file manager
```
Profile files: `soul.md`, `user.md`, `workspace.md`, `memory.md`, `memory/*.md`
## Skills
Skills extend agent functionality via `SKILL.md` files. See [Skills Documentation](./src/agent/skills/README.md).
```bash
multica skills list # List skills
multica skills add owner/repo # Install from GitHub
multica skills status # Check status
```
Built-in: `commit`, `code-review`, `skill-creator`
## Tools
Available tools: `read`, `write`, `edit`, `glob`, `exec`, `process`, `web_fetch`, `web_search`, `memory_search`, `sessions_spawn`
See [Tools Documentation](./src/agent/tools/README.md) for details.
## Architecture ## Architecture
``` ```
@ -202,52 +87,13 @@ Web/Mobile Clients
- **Gateway**: WebSocket server for remote clients - **Gateway**: WebSocket server for remote clients
- **Hub**: Agent lifecycle and event distribution - **Hub**: Agent lifecycle and event distribution
## Time ## Documentation
Super Multica now uses **message-level timestamp injection** for time awareness. | Topic | Link |
Instead of placing dynamic time text in the system prompt, user turns are stamped at runtime. |-------|------|
| Development guide | [docs/development.md](./docs/development.md) |
```mermaid | Credentials & LLM providers | [docs/credentials.md](./docs/credentials.md) |
flowchart TD | CLI usage | [docs/cli.md](./docs/cli.md) |
A[Incoming turn] --> B{Entry point} | Skills & tools | [docs/skills-and-tools.md](./docs/skills-and-tools.md) |
B -->|Desktop/Gateway/Cron/Subagent| C[AsyncAgent.write] | Time injection design | [docs/time-injection.md](./docs/time-injection.md) |
B -->|Heartbeat poll| D[AsyncAgent.write injectTimestamp=false] | Package management | [docs/package-management.md](./docs/package-management.md) |
C --> E{Already stamped or has 'Current time:'?}
E -->|Yes| F[Keep original message]
E -->|No| G[Prefix: [DOW YYYY-MM-DD HH:mm TZ]]
D --> H[Keep original heartbeat prompt]
F --> I[Agent.run]
G --> I
H --> I
I --> J[LLM receives final turn text]
```
### Injection Matrix
| Path | Runtime call | Timestamp injected? | Notes |
| --- | --- | --- | --- |
| Desktop direct chat | `agent.write(content)` | Yes | Default behavior |
| Gateway/remote chat | `agent.write(content)` | Yes | Same entry path as desktop |
| `sessions_spawn` child task | `childAgent.write(task)` | Yes | Child turn gets current time context |
| Cron `agent-turn` payload | `agent.write(cronMessage)` | Yes (guarded) | Skips if message already carries `Current time:` |
| Heartbeat runner | `agent.write(prompt, { injectTimestamp: false })` | No | Prevents heartbeat prompt matching from breaking |
| Internal orchestration | `writeInternal(...)` | No | Uses separate internal run path |
### Why this design
- Keeps system prompt cache-stable (no per-turn date churn in system prompt text)
- Gives the model an explicit "now" reference on each user turn
- Uses guardrails to avoid double-stamping and heartbeat regressions
## Scripts
```bash
pnpm dev # Desktop app (recommended)
pnpm dev:gateway # Gateway only
pnpm dev:web # Web app only
pnpm dev:all # Gateway + Web
pnpm build # Production build
pnpm typecheck # Type check
pnpm test # Run tests
```

30
docs/cli.md Normal file
View file

@ -0,0 +1,30 @@
# CLI
```bash
multica # Interactive mode
multica run "prompt" # Single prompt
multica chat --profile my-agent # Use profile
multica --session abc123 # Continue session
multica session list # List sessions
multica profile list # List profiles
multica skills list # List skills
multica help # Show help
```
Short alias: `mu`
## Sessions
Sessions persist to `~/.super-multica/sessions/<id>/` with JSONL message history and JSON metadata. Context windows are automatically managed with token-aware compaction.
## Profiles
Profiles define agent identity, personality, and memory in `~/.super-multica/agent-profiles/<id>/`.
```bash
multica profile new my-agent # Create profile
multica profile list # List all
multica profile edit my-agent # Open in file manager
```
Profile files: `soul.md`, `user.md`, `workspace.md`, `memory.md`, `memory/*.md`

39
docs/credentials.md Normal file
View file

@ -0,0 +1,39 @@
# Credentials & LLM Providers
## Setup
```bash
multica credentials init
```
Creates:
- `~/.super-multica/credentials.json5` — LLM providers + tools
- `~/.super-multica/skills.env.json5` — skill/plugin API keys
Example `credentials.json5`:
```json5
{
version: 1,
llm: {
provider: "openai",
providers: {
openai: { apiKey: "sk-xxx", model: "gpt-4o" }
}
},
tools: {
brave: { apiKey: "brv-..." }
}
}
```
## LLM Providers
**OAuth Providers** (external CLI login):
- `claude-code` — requires `claude login`
- `openai-codex` — requires `codex login`
**API Key Providers** (configure in `credentials.json5`):
- `anthropic`, `openai`, `kimi-coding`, `google`, `groq`, `mistral`, `xai`, `openrouter`
Check status: `/provider` in interactive mode

82
docs/development.md Normal file
View file

@ -0,0 +1,82 @@
# Development Guide
## Dev Commands
```bash
pnpm dev # Desktop app (recommended)
pnpm dev:desktop # Same as above
pnpm dev:gateway # Gateway only
pnpm dev:web # Web app only
pnpm dev:all # Gateway + Web
pnpm build # Production build (turbo-orchestrated)
pnpm typecheck # Type check all packages
pnpm test # Run tests
pnpm test:watch # Watch mode
pnpm test:coverage # With v8 coverage
```
## Local Full-Stack Development
`pnpm dev:local` starts Gateway + Desktop + Web together with isolated data directories.
**Setup:**
1. Copy `.env.example` to `.env` at the repo root
2. Fill in `TELEGRAM_BOT_TOKEN` (get from [@BotFather](https://t.me/BotFather))
3. Run `pnpm dev:local`
| Service | Address | Notes |
|---------|---------|-------|
| Gateway | `http://localhost:4000` | Telegram long-polling mode |
| Web | `http://localhost:3000` | OAuth login flow |
| Desktop | — | Connects to local Gateway + Web |
Data is stored in `~/.super-multica-dev` and `~/Documents/Multica-dev`, isolated from production.
```bash
pnpm dev:local:archive # Archive dev data and start fresh
```
## Environment Configuration
**Desktop** (`apps/desktop/.env.*`):
| Variable | Description |
|----------|-------------|
| `MAIN_VITE_GATEWAY_URL` | WebSocket Gateway URL for remote device pairing |
| `MAIN_VITE_WEB_URL` | Web app URL for OAuth login redirect |
**Web** (`apps/web/next.config.ts`):
| Variable | Description |
|----------|-------------|
| `API_URL` | Backend API URL (default: `https://api-dev.copilothub.ai`) |
**Build for different environments:**
```bash
# Desktop
pnpm --filter @multica/desktop build # Production (.env.production)
pnpm --filter @multica/desktop build:staging # Staging (.env.staging)
# Web (Vercel)
# Set API_URL in Vercel Dashboard → Settings → Environment Variables
```
See `apps/desktop/.env.example` for the full variable reference.
## Monorepo Workflow
| Command | Purpose |
|---------|---------|
| `pnpm dev` | Full dev mode — watches `core`, `types`, `utils` packages |
| `pnpm dev:desktop` | Desktop only — skip package watching |
**When modifying packages:**
1. Edit code in `packages/core`, `packages/types`, or `packages/utils`
2. Terminal shows `[core] ESM ⚡️ Build success` (~100ms)
3. Restart Desktop to apply changes (Ctrl+C, then `pnpm dev`)
> **Why restart?** Electron main process does not support hot reload — this is an Electron limitation, not ours.

19
docs/skills-and-tools.md Normal file
View file

@ -0,0 +1,19 @@
# Skills & Tools
## Skills
Skills extend agent functionality via `SKILL.md` files. See [Skills Documentation](../packages/core/src/agent/skills/README.md).
```bash
multica skills list # List skills
multica skills add owner/repo # Install from GitHub
multica skills status # Check status
```
Built-in: `commit`, `code-review`, `skill-creator`
## Tools
Available tools: `read`, `write`, `edit`, `glob`, `exec`, `process`, `web_fetch`, `web_search`, `memory_search`, `sessions_spawn`
See [Tools Documentation](../packages/core/src/agent/tools/README.md) for details.

36
docs/time-injection.md Normal file
View file

@ -0,0 +1,36 @@
# Time Injection Design
Super Multica uses **message-level timestamp injection** for time awareness.
Instead of placing dynamic time text in the system prompt, user turns are stamped at runtime.
```mermaid
flowchart TD
A[Incoming turn] --> B{Entry point}
B -->|Desktop/Gateway/Cron/Subagent| C[AsyncAgent.write]
B -->|Heartbeat poll| D[AsyncAgent.write injectTimestamp=false]
C --> E{Already stamped or has 'Current time:'?}
E -->|Yes| F[Keep original message]
E -->|No| G[Prefix: [DOW YYYY-MM-DD HH:mm TZ]]
D --> H[Keep original heartbeat prompt]
F --> I[Agent.run]
G --> I
H --> I
I --> J[LLM receives final turn text]
```
## Injection Matrix
| Path | Runtime call | Timestamp injected? | Notes |
| --- | --- | --- | --- |
| Desktop direct chat | `agent.write(content)` | Yes | Default behavior |
| Gateway/remote chat | `agent.write(content)` | Yes | Same entry path as desktop |
| `sessions_spawn` child task | `childAgent.write(task)` | Yes | Child turn gets current time context |
| Cron `agent-turn` payload | `agent.write(cronMessage)` | Yes (guarded) | Skips if message already carries `Current time:` |
| Heartbeat runner | `agent.write(prompt, { injectTimestamp: false })` | No | Prevents heartbeat prompt matching from breaking |
| Internal orchestration | `writeInternal(...)` | No | Uses separate internal run path |
## Why This Design
- Keeps system prompt cache-stable (no per-turn date churn in system prompt text)
- Gives the model an explicit "now" reference on each user turn
- Uses guardrails to avoid double-stamping and heartbeat regressions