Tron

⚠️ Experimental Project - This project is under active development and changing rapidly. APIs, configuration, and features may change without notice.

🔓 Security Warning - This project is highly insecure. It executes arbitrary plugins, has no authentication beyond Signal's, and should only be run in trusted environments. Do not expose to untrusted users.

A Signal messaging bot with LLM integration and plugin support.

Quick Examples

Once running, chat with the bot naturally. It uses plugins automatically based on your requests:

You: "What tasks do I have pending?"
Bot: [invokes task plugin]
    You have 3 pending tasks:
    1. Call the plumber (due: tomorrow)
    2. Review budget spreadsheet (due: Friday)
    3. Book dentist appointment

You: "Add a task to pick up groceries"
Bot: [invokes task plugin]
    Added: "Pick up groceries"

You: "What's using all my CPU?"
Bot: [invokes ps plugin]
    Top processes by CPU:
    - firefox (12.3%)
    - code (8.1%)
    - slack (4.2%)

You: "Mark task 1 as done"
Bot: [invokes task plugin]
    Completed: "Call the plumber"

In group chats, prefix messages with the trigger keyword (default: T):

You: "T what tasks are due today?"
Bot: [responds to the group]

Prerequisites

• Go 1.25.3+
• signal-cli running in JSON-RPC daemon mode
• An LLM API endpoint (OpenAI-compatible, e.g., DeepInfra, OpenAI, local Ollama)
• SQLite3 (for conversation memory and reminders)
• Taskwarrior (optional, for task plugin)

Setting up signal-cli

1. Install signal-cli from https://github.com/AsamK/signal-cli
2. Register or link your bot account
3. Run in JSON-RPC daemon mode:

signal-cli -a +1234567890 daemon --http=localhost:8080

The bot connects to the JSON-RPC endpoint for sending messages and subscribes to SSE events for receiving them.

Building

# Build the binary
make build

# Build with formatting, vetting, and tests
make all

# Clean build artifacts
make clean

Running

# Source environment config
source .env

# Build and run
make run

# Run with debug logging
make run-debug

Configuration

Configuration can be done via YAML file, environment variables, or both. Environment variables take precedence over YAML values.

Priority: Defaults → YAML file → Environment variables (highest)

YAML Config File

Copy the example config and customize:

cp config.example.yaml config.yaml

Run with the config file:

./bin/tron -config config.yaml

Example config.yaml:

# Signal Configuration
signal_cli_url: "http://localhost:8080"
signal_bot_account: "+1234567890"
signal_operator: "+0987654321"

# LLM Configuration
llm_api_url: "https://api.deepinfra.com/v1/openai"
llm_api_key: "your-api-key-here"
llm_model: "deepseek-ai/DeepSeek-V3.1"
llm_system_prompt: |
  You are a personal assistant bot on Signal.
  Keep responses short. Never use emojis.

# Storage
plugin_dir: "plugins.d"
db_path: "tron.db"

# Behavior
trigger_keyword: "T"
memory_max_messages: 50
memory_max_minutes: 60
daily_summary_hour: 7

Environment Variables

You can also use environment variables (useful for secrets or overriding config):

# Required (if not in YAML)
export SIGNAL_BOT_ACCOUNT="+1234567890"
export SIGNAL_OPERATOR="your-uuid-or-number"
export LLM_API_KEY="your-api-key"

# Optional
export SIGNAL_CLI_URL="http://localhost:8080"
export LLM_API_URL="https://api.deepinfra.com/v1/openai"
export LLM_MODEL="deepseek-ai/DeepSeek-V3.1"
export LLM_SYSTEM_PROMPT="You are a helpful assistant..."
export PLUGIN_DIR="plugins.d"
export DB_PATH="tron.db"
export TRIGGER_KEYWORD="T"
export MEMORY_MAX_MESSAGES="50"
export MEMORY_MAX_MINUTES="60"
export DAILY_SUMMARY_HOUR="7"

Mixed Usage

A common pattern is to put non-sensitive config in YAML and secrets in environment variables:

# Keep API key out of config file
export LLM_API_KEY="your-api-key"

# Run with YAML config (env var overrides any llm_api_key in YAML)
./bin/tron -config config.yaml

Plugins

Tron supports external plugins (shell scripts, Python, etc.) and internal tools (Go-based).

See PLUGINS.md for:

• Using included plugins (task, ps)
• Creating custom plugins
• Plugin configuration

Reminders

🚧 Work in Progress - The reminder system is not fully functional yet. The core scheduling logic exists but may have bugs or missing features.

The reminder system allows scheduling prompts that execute with full tool access. The scheduler checks for due reminders every minute.

Schedule Formats

Format	Example	Description
daily:HH:MM	daily:08:00	Run daily at specified time
hourly:MM	hourly:30	Run every hour at specified minute
interval:DURATION	interval:2h	Run every N duration (30m, 2h, etc.)
cron:EXPR	cron:0 8 * * 1-5	Standard 5-field cron expression
once:DATETIME	once:2024-01-15T08:00	Run once at specified time

Actions

Action	Description
list	Show all configured reminders
add	Create a new reminder (requires: name, prompt, schedule)
delete	Remove a reminder by ID
enable	Enable a paused reminder
disable	Pause a reminder without deleting
run	Execute a reminder immediately

Examples

Creating reminders (via chat):

"Set a reminder to check my tasks every morning at 8am"
→ Creates: daily:08:00 reminder with prompt to list pending tasks

"Remind me every 2 hours to take a break"
→ Creates: interval:2h reminder

"Every weekday at 9am, summarize my calendar"
→ Creates: cron:0 9 * * 1-5 reminder

How it works:

1. The scheduler runs in the background, checking every minute for due reminders
2. When a reminder is due, its prompt is sent to the LLM with full tool access
3. The LLM executes the prompt (can use plugins, check tasks, etc.)
4. The response is sent to the original chat
5. The next run time is calculated based on the schedule type

Reminders are stored in SQLite and persist across restarts.

Makefile Targets

Target	Description
build	Compile binary to bin/tron
run	Build and run
run-debug	Build and run with debug flag
clean	Remove build artifacts
test	Run tests
vet	Run go vet
fmt	Format code
all	fmt, vet, test, build

Usage

Once running, the bot:

• Responds to direct messages from the configured operator
• Responds to group messages prefixed with the trigger keyword (default: T)
• Maintains conversation context per chat
• Sends a daily summary at the configured hour