openlama 0.1.80

Personal AI agent bot — Telegram + Ollama

A fully local AI agent bot powered by Ollama.
Chat via Telegram or terminal — tool calling, image generation, scheduled tasks, custom skills.
All running on your own hardware. Your data never leaves your machine.

한국어

---

Why openlama?

Most AI assistants send your data to cloud servers. openlama runs entirely on your local machine using Ollama, giving you a personal AI agent with full tool access and zero data leakage.

Optimized for Gemma 4 models, but works with any Ollama-compatible model.

---

Features

• 100% Local — No cloud APIs. All processing on your hardware.
• Dual Channel — Telegram bot + terminal TUI with shared conversation context.
• 20+ Built-in Tools — Web search, code execution, file I/O, image generation, Git, Obsidian, and more.
• Custom Skills — Create reusable instruction sets triggered by keywords.
• MCP Support — Connect external tool servers via Model Context Protocol.
• Scheduled Tasks — Cron-based recurring tasks executed by AI.
• Memory System — Two-tier memory: long-term (MEMORY.md) + episodic daily (auto-saved digests).
• Multi-prompt System — SOUL, USERS, MEMORY, SYSTEM prompts for fine-grained personality control.
• Auto-update — openlama update upgrades both openlama and Ollama.
• Cross-platform — macOS, Linux, Windows, Android (Termux).
• Mobile Device Control — On Android, control your phone: camera, SMS, location, sensors, and more via Termux:API.
• Self-healing — openlama doctor fix auto-diagnoses and repairs issues.

---

Quick Start

1. Install

# Recommended
uv tool install openlama

# Or with pip
pip install openlama

2. Setup

openlama setup

The interactive wizard will:

  ● Step 1/7 — Ollama
  ✓ Ollama is installed
  ✓ Ollama server running (v0.20.3)

  ● Step 2/7 — Models
  ? Select models to download:
    ✓ gemma4:e4b       9.6 GB  [recommended]
      qwen3:8b         5.2 GB  [light]
      deepseek-r1:8b   5.2 GB  [coding]
    ✓ gemma4:e2b                [installed]

  gemma4:e4b (pulling manifest)  ━━━━━━━━━━━━━━  4.2/9.6 GB  52.3 MB/s  0:01:43

  ● Step 3/7 — Channel
  ? Enter Telegram bot token (@BotFather): 1234567890:ABC...
  ✓ Connected: @your_bot_name

  ● Step 4/7 — Password
  ? Set admin password: ********

  ● Step 5/7 — Features
  ✓ ComfyUI detected: macOS Desktop App

  ● Step 6/7 — Voice Recognition (STT)
  ✓ faster-whisper is installed

  ● Step 7/7 — Obsidian Notes
  ✓ obsidian-cli is installed
  ✓ Vault connected: 13 items found

  ╭─────────────────────────────────────────────╮
  │  ✅ Setup complete!                          │
  │                                              │
  │  Start:   openlama start                     │
  │  Chat:    openlama chat                      │
  │  Doctor:  openlama doctor                    │
  ╰─────────────────────────────────────────────╯

3. Run

# Start Telegram bot in background
openlama start -d

# Open terminal chat (shares context with Telegram)
openlama chat

4. Health Check

openlama doctor

  ✓  Data directory         /home/user/.config/openlama
  ✓  Database               7 tables
  ✓  Telegram bot token     Set (12345678...nqbw)
  ✓  Python dependencies    All critical packages available
  ✓  Boot service           systemd user service registered
  ✓  Disk space             120.5 GB free
  ✓  Telegram connection    Bot @your_bot is reachable
  ✓  Ollama server          Connected (http://127.0.0.1:11434)
  ✓  Ollama version         v0.20.3 (latest)
  ✓  Ollama models          gemma4:e4b, gemma4:e2b
  !  ComfyUI                Not running (auto-start configured)

  17 passed · 1 warning(s)

---

Terminal Chat (TUI)

openlama chat

──────────────────────────── openlama ─────────────────────────────
  model: gemma4:e4b | ctx: 12% (8 turns) | telegram: @your_bot
  Type / for commands, /quit to exit.

You: What's the weather in Seoul?

╭──────────────────────────── AI ─────────────────────────────────╮
│                                                                  │
│  I'll search for that.                                           │
│                                                                  │
│  Based on current data, Seoul is 18°C with partly cloudy skies.  │
│  Humidity is 45% with light winds from the northwest.            │
│                                                                  │
╰──────────────────────────────────────────────────────────────────╯
  📊 ██░░░░░░░░░░░░░░░░░░ 12.3% (2,841/32,768 tokens)  |  turns: 9

Chat Commands

Type / to see all available commands:

  Chat
    /help             Show available commands
    /clear            Clear conversation context
    /status           Show session and context info
    /compress         Compress conversation context
    /session          View/extend session
    /export           Export conversation history
    /profile          Redo profile setup
    /quit             Exit chat

  Model
    /model            Show or change current model
    /models           List available models (with capabilities)
    /pull             Download a new model
    /rm               Delete a model

  Settings
    /settings         Interactive model settings
    /set <p> <v>      Change a parameter
    /think            Toggle think/reasoning mode
    /systemprompt     View/edit prompt files

  System
    /ollama           Ollama server management
    /skills           List installed skills
    /mcp              MCP server status
    /cron             View and manage scheduled tasks

---

Telegram Bot

After openlama start, open your bot in Telegram:

1. Login — Send any message, enter the admin password
2. Profile Setup — Select language, describe yourself, set agent identity
3. Chat — Start chatting. The bot uses all available tools automatically.

Telegram Features

• Inline keyboard menus for settings, model selection
• Streaming responses with real-time edits
• Image/document/audio/video/ZIP analysis
• Voice message transcription (STT via faster-whisper)
• Context bar showing token usage (Ollama actual tokens)
• Prompt file editor via inline buttons

---

Built-in Tools (20+)

Tool	Description
web_search	Search the web via DuckDuckGo
url_fetch	Fetch and extract text from URLs
calculator	Evaluate math expressions
code_execute	Run Python, Node.js, or Shell code
shell_command	Execute system commands
file_read	Read files or list directories
file_write	Write or append to files
git	Git operations (status, log, diff, commit)
process_manager	List/kill processes, system status
tmux	Full tmux terminal multiplexer control
image_generate	Text-to-image via ComfyUI
image_edit	Image editing via ComfyUI
memory	Two-tier memory: long-term + daily episodic
skill_creator	Create/manage/install custom skills
mcp_manager	Install/manage MCP tool servers
cron_manager	Schedule recurring AI tasks
get_datetime	Current date and time
self_update	Check and install openlama updates
whisper	Audio/voice transcription (STT, optional)
obsidian	Obsidian vault read/write/search (optional)
termux_device	Android device control via Termux:API (Android only)

The AI understands tool requests in any language:

"서버 상태 확인해줘" → shell_command "search for latest AI news" → web_search "매일 10시에 뉴스 요약해줘" → cron_manager "노트 목록 보여줘" → obsidian "배터리 확인해줘" → termux_device (Android)

---

Android (Termux) Setup

openlama runs on Android via Termux. Two modes are supported:

Mode 1: Remote Inference (Recommended)

Run the bot on your phone, inference on a desktop/server with a GPU.

Prerequisites

• Termux — Install from F-Droid or GitHub Releases (recommended). The Google Play version works for basic bot operation but lacks plugin support (see note below).
• Termux:API — For full device control (camera, SMS, GPS, sensors). F-Droid/GitHub only.
• Termux:Boot — For auto-start on boot. F-Droid/GitHub only.
• A desktop/server running Ollama (accessible on the network)

F-Droid vs Play Store vs GitHub:

	F-Droid / GitHub	Google Play
Bot daemon + remote Ollama	✅	✅
Termux:API plugin (35 device actions)	✅	❌ (some built-in)
Termux:Boot (auto-start on boot)	✅	❌
Latest features (v0.118+)	✅	❌ (equivalent to v0.108)

All Termux APKs must come from the same source (F-Droid, GitHub, or Play Store). Mixing sources will fail due to different signing keys. F-Droid and GitHub APKs share the same key and are interchangeable.

Google Play Protect may block F-Droid/GitHub APK installation. Dismiss the warning or temporarily disable Play Protect during install.

Installation

# 1. Update Termux packages
pkg update && pkg upgrade -y

# 2. Install Python and Termux:API bridge
pkg install python termux-api -y

# 3. Install openlama
pip install openlama

# 4. Run setup wizard
openlama setup
#   Step 1: Select "Remote" → enter server URL (e.g., http://192.168.1.100:11434)
#   Step 2: Select model from remote server
#   Step 3: Enter Telegram bot token
#   Step 4: Set password

# 5. Start the bot
openlama start -d

# 6. (Optional) Auto-start on boot (F-Droid/GitHub only)
openlama start --install-service

Note: On the remote Ollama server, start with OLLAMA_HOST=0.0.0.0 ollama serve to accept network connections.

Mode 2: On-Device Inference

Run everything on the phone (requires 8GB+ RAM).

# Install Ollama via Termux User Repository
pkg install tur-repo -y
pkg install ollama python termux-api -y

# Install openlama and run setup
pip install openlama
openlama setup    # Select "Local" → downloads a model (~3-7 GB)

openlama start -d

Android Device Control

When running on Android, the termux_device tool gives the AI control over your phone (requires Termux:API from F-Droid/GitHub):

Category	Actions
Phone	call, sms_send, sms_list, call_log, contacts
Camera	camera_photo (front/rear), camera_info
Audio	mic_record, media_play, tts_speak, volume_get/set
Sensors	location, battery, sensor_list/read
System	brightness, torch, clipboard, wifi_info/scan
Notifications	notification, toast, vibrate
Apps	app_launch, app_list, share, download

Safety rules are enforced:

• Phone calls and SMS require explicit user confirmation
• Location data is never shared without consent

Mobile Recommended Models

Model	Size	Notes
gemma4:e2b	7.2 GB	Best for mobile — 2.3B effective params
gemma3:4b	3.3 GB	Good balance
phi4-mini	2.5 GB	Lightweight
gemma3:1b	0.8 GB	Ultra-light, minimal hardware

Keeping openlama Alive on Android

openlama acquires a wake lock automatically to keep the CPU running when the screen is off. However, wake lock alone is not sufficient on modern Android — you must also configure your device:

Required (all devices):

• Disable battery optimization: Settings → Apps → Termux → Battery → Unrestricted

Required (Android 12+):

• Disable phantom process killer: Settings → Developer Options → Disable child process restrictions
• If Developer Options is not available, enable it via Settings → About Phone → tap Build Number 7 times

OEM-specific (check dontkillmyapp.com for your device):

• Samsung: Settings → Battery → Background usage limits → Never sleeping apps → Add Termux
• Xiaomi/MIUI: Settings → Battery → App battery saver → Termux → No restrictions; also enable Auto-start
• Huawei/EMUI: Settings → Battery → App launch → Termux → Manage manually (enable all)
• OnePlus: Settings → Battery → Battery optimization → Termux → Don't optimize

---

Memory System

openlama uses a two-tier memory architecture:

Long-term Memory (MEMORY.md)

• Stores important facts, user preferences, key decisions.
• Managed via the memory tool (save/list/search/delete).
• Accessed by keyword search — not loaded into system prompt to save context for local LLMs.

Episodic Daily Memory (memories/YYYY-MM-DD.md)

• Auto-saved conversation digests on context compression, clear, and daily flush.
• Searchable by date and keyword via the memory tool (list_dates/read_daily/search_daily).
• Enables the AI to recall past conversations: "What did we talk about yesterday?"

---

Custom Skills

Skills are reusable instruction sets that activate on trigger keywords.

Create via CLI

openlama skill create

Create via Chat

"Create a skill called 'code-reviewer' that triggers when I say 'review this' — it should read the file, check for bugs, and suggest fixes"

Skill File Format

~/.config/openlama/skills/<name>/SKILL.md:

---
name: code-reviewer
description: "Activated when user asks for code review"
trigger: "review, code review, check this code"
---

## Rules
1. Read the file specified by the user
2. Check for bugs, security issues, performance problems
3. Suggest improvements with code examples

---

MCP Integration

Connect external tools via Model Context Protocol:

# Add a server
openlama mcp add github npx -y @github/github-mcp

# With environment variables
openlama mcp add github npx -y @github/github-mcp -e GITHUB_TOKEN=ghp_xxx

# List servers
openlama mcp list

# Remove
openlama mcp remove github

MCP tools are automatically registered and available to the AI.

---

Scheduled Tasks

Natural language scheduling — the AI converts to cron expressions:

"Check disk usage every hour" → 0 */1 * * * "Summarize tech news every day at 9am" → 0 9 * * * "Monitor server health every 5 minutes" → */5 * * * *

Each execution is a one-shot AI call with full tool access. Results are sent to your chat.

openlama cron list       # View all tasks
openlama cron delete 1   # Remove a task

---

Prompt System

openlama uses a multi-file prompt architecture:

File	Purpose	Editable
SYSTEM.md	Tools, rules, skills list	Auto-generated each request
SOUL.md	Agent identity and personality	Yes — /systemprompt
USERS.md	User profile and language	Yes — /systemprompt
MEMORY.md	Long-term memory entries	Via memory tool (not in prompt)

All files are in ~/.config/openlama/prompts/ and can be edited via:

• Telegram: /systemprompt → select file → edit → send back
• CLI: /systemprompt → opens in $EDITOR (nano/vim/code)

---

Architecture

~/.config/openlama/
├── openlama.db              # SQLite (settings, users, context, cron jobs)
├── openlama.pid             # Daemon PID file
├── openlama.log             # Daemon log
├── mcp.json                 # MCP server configuration
├── prompts/
│   ├── SYSTEM.md            # Auto-generated system prompt
│   ├── SOUL.md              # Agent identity
│   ├── USERS.md             # User profile
│   └── MEMORY.md            # Long-term memory (tool-accessed only)
├── memories/
│   └── YYYY-MM-DD.md        # Episodic daily memory
├── skills/
│   └── <name>/SKILL.md      # Custom skills
└── workflows/
    ├── txt2img_default.json  # ComfyUI text-to-image
    └── img2img_default.json  # ComfyUI image-to-image

---

CLI Reference

Command	Description
openlama setup	Interactive setup wizard
openlama start	Start Telegram bot (foreground)
openlama start -d	Start as background daemon
openlama start --install-service	Register OS auto-start service
openlama start --uninstall-service	Remove OS auto-start service
openlama stop	Stop daemon
openlama restart	Restart daemon
openlama chat	Terminal chat TUI
openlama status	Connection and process status
openlama doctor	Run diagnostic checks
openlama doctor fix	Auto-fix detected issues
openlama update	Update openlama + Ollama
openlama config list	View all settings
openlama config get <key>	Get a setting value
openlama config set <key> <value>	Change a setting (auto-restarts daemon)
openlama config reset	Reset all settings
openlama config stt	Show STT status
openlama config stt install	Install faster-whisper for voice recognition
openlama config stt enable/disable	Enable/disable STT
openlama config obsidian	Show Obsidian integration status
openlama config obsidian install	Install obsidian-cli
openlama config obsidian vault <name>	Set Obsidian vault
openlama config obsidian disable	Disable Obsidian integration
openlama skill list	List installed skills
openlama skill create	Create a new skill interactively
openlama skill delete <name>	Delete a skill
openlama mcp list	List MCP servers
openlama mcp add <name> <cmd> [args]	Add an MCP server
openlama mcp remove <name>	Remove an MCP server
openlama tool list	List all registered tools
openlama cron list	List scheduled tasks
openlama cron delete <id>	Delete a scheduled task
openlama logs	View daemon logs
openlama --version	Show version

---

Recommended Models

Desktop / Server

Model	Size	Best For
gemma4:e4b	9.6 GB	Overall best — recommended default
gemma3:4b	3.3 GB	Fast responses, lower memory
qwen3.5:4b	3.4 GB	Good multilingual support
qwen3:8b	5.2 GB	Strong reasoning
deepseek-r1:8b	5.2 GB	Coding tasks
gemma3:1b	0.8 GB	Ultra-light, minimal hardware

Mobile (Android)

Model	Size	Best For
gemma4:e2b	7.2 GB	Best for mobile — 2.3B effective params
gemma3:4b	3.3 GB	Good balance for mobile
phi4-mini	2.5 GB	Lightweight
gemma3:1b	0.8 GB	Ultra-light, 1GB RAM devices

---

System Requirements

Desktop / Server

Component	Minimum	Recommended
Python	3.11+	3.13+
RAM	4 GB	8 GB+
Disk	5 GB	20 GB+ (for models)
OS	macOS / Linux / Windows	macOS (Apple Silicon)
Ollama	Required	Latest version
ComfyUI	Optional	For image generation

Android (Termux)

Component	Minimum	Recommended
Android	7+	12+
RAM	4 GB (remote mode)	8 GB+ (on-device)
Disk	500 MB (remote)	8 GB+ (on-device)
Termux	Required	From F-Droid or GitHub
Termux:API	Recommended	For device control (F-Droid/GitHub only)
Termux:Boot	Optional	For auto-start (F-Droid/GitHub only)

---

Configuration

All settings are stored in SQLite (~/.config/openlama/openlama.db).

Override the data directory:

export OPENLAMA_DATA_DIR=/custom/path

Key settings:

Key	Default	Description
telegram_bot_token	—	Telegram bot API token
default_model	—	Default Ollama model
ollama_base	http://127.0.0.1:11434	Ollama API URL
comfy_enabled	false	Enable ComfyUI integration
comfy_base	http://127.0.0.1:8184	ComfyUI API URL
tool_sandbox_path	~/sandbox	Sandbox for code execution
obsidian_vault	—	Obsidian vault name (enables obsidian tool)
stt_enabled	auto	Voice recognition: true/false/auto

---

Contributing

Contributions are welcome! Please:

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Make your changes
4. Run tests (pytest)
5. Commit (git commit -m 'feat: add amazing feature')
6. Push (git push origin feature/amazing-feature)
7. Open a Pull Request

Development Setup

git clone https://github.com/sussa3007/openlama.git
cd openlama
uv venv && source .venv/bin/activate
uv pip install -e ".[dev]"
openlama setup

---

Roadmap

• Web UI channel
• Discord channel
• iOS Shortcuts integration
• Multi-user with separate contexts
• RAG (Retrieval-Augmented Generation) with local documents
• Voice input/output
• Plugin marketplace

---

License

MIT

---

Built with Ollama, python-telegram-bot, Rich, and Click.
_{Your AI, your hardware, your data.}