███████╗██████╗ ██╗      █████╗ ███████╗██╗  ██╗
   ██╔════╝██╔══██╗██║     ██╔══██╗██╔════╝██║  ██║
   ███████╗██████╔╝██║     ███████║███████╗███████║
   ╚════██║██╔═══╝ ██║     ██╔══██║╚════██║██╔══██║
   ███████║██║     ███████╗██║  ██║███████║██║  ██║
   ╚══════╝╚═╝     ╚══════╝╚═╝  ╚═╝╚══════╝╚═╝  ╚═╝

Splash is a local-first autonomous agent you call from any directory. One global config file, one command — splash — and it builds, edits, debugs, runs shells, reads the web, drives bots, whatever the task needs.

Under the hood: a 10-phase agent loop (intake → understand → plan → context → tool-select → execute → verify → correct → finalize → persist) over a multi-provider router (OpenRouter/Kimi, Claude, GPT, Gemini, DeepSeek, Ollama), backed by a persistent run store you can inspect live in an Ink TUI.

It's designed to feel like OpenClaw / NanoClaw — install it, drop a key, go.

# 1. Install
git clone https://github.com/AzizX-coder/Splash.git && cd Splash
pnpm install
pnpm link --global          # gives you the `splash` command anywhere

# 2. Configure (30 seconds)
splash init                 # pick provider, paste key, pick model, pick safety

# 3. Go
splash "summarize this folder"
splash "build a FastAPI todo service" --background
splash tui                  # live dashboard

No pnpm dev. No per-project .env. Just splash.

splash telegram · splash slack · splash whatsapp · splash messenger · splash discord

Each reads its credentials from ~/.splash/config.json (or env). Webhook ports: Slack 3001, WhatsApp 3002, Messenger 3003, Discord 3004. Telegram uses long-poll.

┌─ 💧 Splash · Autonomous Agent Dashboard ──── ≈ ─┐
│                                                   │
│ ┌── Runs ───┐ ┌── Task Graph ────────────────┐   │
│ │ ▸ ◌ build │ │ ✓ intake  ✓ plan  ◉ execute  │   │
│ │   ✓ test  │ │ ○ verify  ○ finalize         │   │
│ │   ✗ scan  │ └──────────────────────────────┘   │
│ │           │ ┌── Resources ─────────────────┐   │
│ │           │ │ steps 4 · tools 12 · 8.3s    │   │
│ │           │ └──────────────────────────────┘   │
│ │           │ ┌── Live Logs ─────────────────┐   │
│ │           │ │ 14:22:11 phase → execute     │   │
│ │           │ │ 14:22:12 ⚡ fs.read_file     │   │
│ │           │ │ 14:22:14 ⚡ terminal.run     │   │
│ │           │ └──────────────────────────────┘   │
│ └───────────┘                                     │
│ ↑/↓ select · s stop · r retry · p pause · q quit │
└───────────────────────────────────────────────────┘

Shortcuts: ↑/↓ select · s stop · r retry · p pause/resume · l reload logs · q quit

bin/splash.mjs                   ← global launcher (resolves tsx, preserves cwd)
examples/cli.ts                  ← CLI router: chat | tui | runs | config | platforms
packages/
  utils/         Result<T,E>, logger, theme (banner, spinners, droplet frames)
  config/        schema + loader + ~/.splash store
  safety/        pattern-based policy engine
  memory/        file-backed persistent memory
  providers/     Claude / OpenAI / Gemini / Ollama / OpenRouter router
  connectors/    fs / terminal / database / http / browser / git
  skills/        18 reusable task recipes
  core/
    agent-loop.ts             10-phase orchestrator
    runs/
      events.ts               typed RunEvent union
      run-store.ts            ~/.splash/runs + ndjson logs (schema v1)
      run-manager.ts          lifecycle + background worker
    tui/app.tsx               Ink dashboard
bots/                          telegram / slack / whatsapp / messenger / discord

Run lifecycle: queued → running → succeeded | failed | cancelled

Each transition is written to the persistent store. Events stream via append-only ndjson so both the CLI (runs logs --follow) and the TUI subscribe to the same source.

Config precedence (lowest → highest):

1. Schema defaults
2. ~/.splash/config.json
3. .env in cwd + SPLASH_* / ALPCLAW_* env vars
4. Explicit loadConfig() overrides

Env var names: SPLASH_DEFAULT_PROVIDER, SPLASH_DEFAULT_MODEL, SPLASH_SAFETY_MODE, SPLASH_MEMORY_PATH, SPLASH_LOG_LEVEL.

Provider keys: ANTHROPIC_API_KEY, OPENAI_API_KEY, OPENROUTER_API_KEY, GOOGLE_API_KEY, DEEPSEEK_API_KEY, OLLAMA_BASE_URL.

splash config preset fast       # permissive safety, Kimi K2, foreground
splash config preset balanced   # standard safety (default)
splash config preset safe       # strict safety, confirms every action

splash config doctor

Checks:

• ✓ Provider key present
• ✓ Config dir writable
• ✓ Runs dir writable
• ✓ Default provider reachable

Prints fix: hints for each failing check. Add --json for automation-friendly output.

# 1. Start a task in the background
splash "refactor auth module to use JWT" --background
# → run_abc123 started (background)

# 2. Check status
splash runs list
# → run_abc123  running  "refactor auth module..."  12s

# 3. Attach in the TUI
splash tui
# → live dashboard with logs, task graph, controls

pnpm install                # install all workspace deps
pnpm test                   # vitest — 93 tests across 12 test files
pnpm -r run build           # type-check + build all 8 packages
pnpm dev                    # same as `splash` without the global link

Goals: Production-grade. Local-first. Fast first-run. Observable. Extensible (drop a new skill or connector in one file).

Non-goals (this phase):

• Multi-host distributed orchestration
• Cloud-hosted run registry
• Web UI

Splash is the rebrand of AlpClaw. The alpclaw binary and ALPCLAW_* env vars still work — they delegate to splash. Legacy ~/.alpclaw/config.json is auto-migrated on first read.

MIT