salmalm 0.27.36

Personal AI Gateway — Python AI assistant with multi-model routing, web UI, and 66 built-in tools. Core: uvicorn + fastapi + websockets.

😈 SalmAlm (삶앎)

Your Entire AI Life in One pip install

한국어 README · Documentation

---

What is SalmAlm?

SalmAlm is a self-hosted personal AI gateway — one Python package that gives you a full-featured AI assistant with web UI, Telegram/Discord bots, 66 tools, memory system, sub-agents, and multi-provider model routing.

No Docker. No Node.js. No config files. Just:

pip install salmalm
python3 -m salmalm
# → http://localhost:18800

First launch opens a Setup Wizard — paste an API key, pick a model, done.

---

⚡ Quick Start

# Install (recommended)
pipx install salmalm

# Or with pip (in a venv)
python3 -m venv ~/.salmalm-env && ~/.salmalm-env/bin/pip install salmalm

# Run
salmalm --open
# → Browser opens http://localhost:18800

# Setup Wizard appears → paste API key → done!

Supported Providers

Provider	Models	Tier
Anthropic	Claude Opus 4.6, Sonnet 4.6, Haiku 4.5	Complex / Moderate / Simple
OpenAI	GPT-5.2, GPT-5.1, o3	Complex / Moderate
Google	Gemini 3 Pro, 3 Flash	Moderate / Simple
xAI	Grok-4, Grok-3-mini	Complex / Simple
Local	Ollama / LM Studio / vLLM	Auto-detected

---

🧠 Architecture

Browser ──SSE/WS──► SalmAlm ──► Anthropic / OpenAI / Google / xAI / Ollama
Telegram ──►           ├── 3-Tier Model Router (simple/moderate/complex)
Discord  ──►           ├── Engine Pipeline (classify → route → context → execute)
                       ├── Memory (MEMORY.md + daily logs + auto-recall + RAG)
                       ├── Sub-Agent Manager (spawn/steer/kill/notify)
                       ├── 62 Tools (risk-tiered, dynamic loading)
                       ├── Vault (PBKDF2-200K + AES-256-GCM)
                       └── Cron / Backup / Self-Diagnostics

---

🎯 Features

AI Engine

• 3-tier auto-routing — simple→Haiku, moderate→Sonnet, complex→Opus/GPT-5 (cost-optimized)
• Extended Thinking — 4 levels (low/medium/high/xhigh) with budget control
• 5-stage context compaction — keeps conversations going without losing context
• Prompt caching — Anthropic cache_control for cost reduction
• Model failover — automatic retry across providers
• Tier momentum — prevents model downgrade mid-complex-task

Memory System

• 2-layer — MEMORY.md (curated long-term) + memory/YYYY-MM-DD.md (daily logs)
• Auto-recall — searches memory before each response, injects relevant context
• Auto-curation — promotes important daily entries to long-term memory
• TF-IDF RAG — cosine similarity search across all files

Sub-Agents

• Spawn background AI workers with independent sessions
• Thinking level per agent, labels, mid-task steering
• Auto-notify on completion (WebSocket + Telegram)

62 Built-in Tools

Shell exec, file I/O, web search (Brave), web fetch, Python eval (opt-in), image gen (DALL-E), TTS/STT, browser automation (Playwright), RAG search, cron jobs, system monitor, and more.

Unique Features

Feature	What it does
Self-Evolving Prompt	AI auto-generates personality rules from conversations (max 20, FIFO)
Dead Man's Switch	Automated actions (email, commands) if you go inactive for N days
Shadow Mode	AI learns your communication style, can reply as you when away
Life Dashboard	Unified view: expenses, habits, calendar, mood, routines
Mood-Aware Response	Detects emotional state from NLP signals, adjusts tone
A/B Split Response	Two model perspectives on the same question, side-by-side
Time Capsule	Schedule encrypted messages to your future self
Thought Stream	Private journaling with hashtag search and mood tracking
Agent-to-Agent	HMAC-SHA256 signed communication between SalmAlm instances
Workflow Engine	Multi-step AI workflows with conditions and loops
Message Queue	5 modes: collect, steer, followup, steer-backlog, interrupt
MCP Marketplace	Install/manage Model Context Protocol tool servers

Web UI

• SSE streaming with real-time thinking display
• Multi-file upload (drag-drop, paste, clip button)
• Session management (branch, rollback, search)
• Command palette (Ctrl+Shift+P), dark/light themes, EN/KR i18n
• Settings: Engine, Routing, Channels, Memory, Cron, Backup
• PWA installable

Channels

• Web — SPA at localhost:18800
• Telegram — polling + webhook with inline buttons
• Discord — bot with thread support

---

💰 Cost Optimization

Technique	Effect
3-tier auto-routing	Simple→$1/M, Complex→$3/M
Dynamic tool loading	62 → 0-12 tools per request
Tool schema compression	91% token reduction
Intent-based max_tokens	Chat 512, code 4096
Response caching	Same question cached 30min-24h

$7/day → $1.2/day at 100 calls/day (83% savings)

---

🔒 Security

All dangerous features default OFF:

Feature	Default	Opt-in
Network bind	127.0.0.1	SALMALM_BIND=0.0.0.0
Shell operators	Blocked	SALMALM_ALLOW_SHELL=1
Python eval	Disabled	SALMALM_PYTHON_EVAL=1

Plus: SSRF defense, CSRF protection, CSP, audit logging, memory scrubbing, path validation, 150+ security tests.

See SECURITY.md for details.

---

🦙 Local LLM

# Ollama
ollama serve
# → Settings → Local LLM → http://localhost:11434/v1 → models auto-discovered

Also supports LM Studio (localhost:1234/v1) and vLLM (localhost:8000/v1).

---

📊 Codebase

Metric	Value
Python files	192
Lines of code	~52,760
Tools	62
Tests	1,908 passing
Max cyclomatic complexity	≤20 (all but 1 function)
Files > 800 lines	0

---

🤝 Contributing

git clone https://github.com/hyunjun6928-netizen/salmalm.git
cd salmalm
pip install -e ".[dev]"
python -m pytest tests/ -q --timeout=30 -x \
  --ignore=tests/test_multi_tenant.py \
  --ignore=tests/test_fresh_install_e2e.py

---

📄 License

MIT

---

SalmAlm = 삶(Life) + 앎(Knowledge)

Your life, understood by AI.