LXMFy

Python framework for LXMF bots on the Reticulum Network.

Docs

Features

Category	Capabilities
Core	Interactive CLI, command prefixes, cron-style task scheduler, middleware and event systems
Connectivity	Direct delivery with propagation fallback, auto-peering, RNS links, opportunistic sending, RRC hub client
Security	Spam protection, role-based permissions, identity pinning, message signature policy, Landlock LSM filesystem sandbox on Linux
NLP	Local offline intent classification, type-hinted argument parsing
Extensions	Python cogs, external script cogs (Bash, Go, C, and others), subprocess sandboxing via Landlock, bwrap, or firejail
Storage	JSON, SQLite, and in-memory backends, crash-safe outgoing message persistence
Reliability	Stability and stress tests, chaos engineering hooks, resource leak checks
UX	Help on first message, auto-generated help menus, customizable bot icons, attachments

Installation

Requirements: Python 3.11+, RNS 1.4.2+, LXMF 1.1.1+, cbor2 5.4.0+ (pulled in with LXMFy).

From PyPI

# pip
pip install lxmfy

# pipx
pipx install lxmfy

From source

git clone https://git.quad4.io/LXMFy/LXMFy.git
cd LXMFy
poetry install

Usage

lxmfy

Create a bot project:

lxmfy create

Debug send and receive:

lxmfy debug
lxmfy debug --config ./config --output ./lxmfy-debug-report.txt
lxmfy debug probe <destination_hash> --request-path --wait 30
lxmfy debug send <destination_hash>
lxmfy debug receive
lxmfy debug compare <hash_a> <hash_b>

Doctor mode prints a verdict and next steps, then categorized checks (OS, shared vs owned instance, disk permissions, interfaces, announce, send pipeline / storage history, receive readiness). It saves a privacy-redacted lxmfy-debug-*.txt you can share. Colors turn off when stdout is not a TTY, when NO_COLOR is set, or when Windows VT is unavailable. Use --no-color or NO_COLOR=1 for plain output.

Docker

Build and run

From the project root:

docker build -t lxmfy-test .

docker run -d \
    --name lxmfy-test-bot \
    -v $(pwd)/config:/bot/config \
    -v $(pwd)/.reticulum:/root/.reticulum \
    --restart unless-stopped \
    lxmfy-test

Host networking (AutoInterface):

docker run -d \
    --name lxmfy-test-bot \
    --network host \
    -v $(pwd)/config:/bot/config \
    -v $(pwd)/.reticulum:/root/.reticulum \
    --restart unless-stopped \
    lxmfy-test

Build a wheel

docker build -f docker/Dockerfile.Build -t lxmfy-wheel-builder .
docker run --rm -v "$(pwd)/dist_output:/output" lxmfy-wheel-builder

That copies the built wheel into ./dist_output.

Example

from lxmfy import LXMFBot, load_cogs_from_directory

bot = LXMFBot(
    name="LXMFy Test Bot", # Name of the bot that appears on the network.
    announce=5400, # Announce every hour, set to 0 to disable.
    announce_enabled=True, # Set to False to disable all announces (both initial and periodic)
    announce_immediately=True, # Set to False to disable initial announce
    admins=["your_lxmf_hash_here"], # List of admin hashes.
    hot_reloading=True, # Enable hot reloading.
    command_prefix="/", # Set to None to process all messages as commands.
    cogs_dir="cogs", # Specify cogs directory name.
    rate_limit=5, # 5 messages per minute
    cooldown=5, # 5 seconds cooldown
    max_warnings=3, # 3 warnings before ban
    warning_timeout=300, # Warnings reset after 5 minutes
    signature_verification_enabled=True, # Enable cryptographic signature verification
    require_message_signatures=False, # Allow unsigned messages but log them
    propagation_fallback_enabled=True, # Enable propagation fallback after direct delivery fails
    propagation_node="your_propagation_node_hash_here", # Manual propagation node (optional)
    autopeer_propagation=True, # Auto-discover propagation nodes (optional)
    autopeer_maxdepth=4, # Max hops for auto-peering (default: 4)
    enable_propagation_node=False, # Run as propagation node (default: False)
    message_storage_limit_mb=500, # Storage limit in MB for propagation node (default: 500)
    direct_delivery_retries=3, # Number of direct delivery attempts before falling back to propagation
    landlock_enabled=True, # Linux Landlock LSM sandbox for the bot process (default)
    external_cogs_sandbox_enabled=True, # Sandbox external script cogs on Linux
    external_cogs_sandbox_type="auto", # auto, landlock, bwrap, firejail, or none
)

load_cogs_from_directory(bot)

@bot.command(name="ping", description="Test if bot is responsive")
def ping(ctx):
    ctx.reply("Pong!")

@bot.command(name="echo", description="Echo a message", admin_only=True)
def echo(ctx, message: str):
    ctx.reply(message)

bot.run()

RRC (Reticulum Relay Chat)

Bots can join RRC hubs as ordinary clients over RNS Links with CBOR envelopes:

from lxmfy import LXMFBot, RRCMessage

bot = LXMFBot(
    name="RoomBot",
    rrc_enabled=True,
    rrc_hubs=["your_rrc_hub_destination_hash"],
    rrc_rooms=["lobby"],
    rrc_nick="RoomBot",
)

@bot.on_rrc
def on_rrc(event, client, payload):
    if event == "msg" and isinstance(payload, RRCMessage) and payload.mention:
        client.send_message(payload.room, f"Heard you, {payload.nick}")

bot.run()

Or connect at runtime with bot.connect_rrc(hub_hash, rooms=["lobby"]).

Hub sessions persist across restarts by default (rrc_persist_sessions=True). Outgoing LXMF messages are also persisted by default (message_persistence_enabled=True) so a crash mid-queue does not drop them. The outbound queue is bounded (message_queue_size, default 50) and drops the oldest message when full.

Propagation Node Configuration

LXMFy supports three modes for propagation node usage.

Manual configuration

Set a specific propagation node by hash:

bot = LXMFBot(
    name="MyBot",
    propagation_fallback_enabled=True,
    propagation_node="your_propagation_node_hash_here",
    direct_delivery_retries=3,
)

Automatic discovery (auto-peering)

Discover propagation nodes from network announces:

bot = LXMFBot(
    name="MyBot",
    propagation_fallback_enabled=True,
    autopeer_propagation=True,
    autopeer_maxdepth=4,
)

The bot peers with suitable nodes within autopeer_maxdepth hops.

Run as a propagation node

Store and forward messages for offline recipients:

bot = LXMFBot(
    name="MyPropagationBot",
    enable_propagation_node=True,
    message_storage_limit_mb=500,
)

message_storage_limit_mb caps disk use. Set to 0 for unlimited storage (not recommended).

Querying propagation status

status = bot.get_propagation_node_status()
print(f"Current outbound node: {status['current_outbound_node']}")
print(f"Discovered peers: {status['discovered_peers']}")

Setting the propagation node at runtime

bot.set_propagation_node("new_propagation_node_hash")

Storage limits

stats = bot.get_propagation_storage_stats()
print(f"Storage used: {stats['storage_size_mb']:.2f} MB")
print(f"Storage limit: {stats['storage_limit_mb']} MB")
print(f"Utilization: {stats['utilization_percent']:.1f}%")
print(f"Messages stored: {stats['message_count']}")

bot.set_message_storage_limit(megabytes=1000)

Propagation notes

• Without manual config, auto-peering, or running as a node, messages that need propagation fail
• You can combine modes (manual node plus auto-peering as backup)
• A propagation-node bot still sends and receives normally
• Auto-peering respects autopeer_maxdepth so distant nodes are skipped

Security and Sandboxing

On Linux kernels with Landlock support (5.13+), LXMFy can restrict filesystem access for the bot process and for external script cogs.

Bot process sandbox

When landlock_enabled=True (default), the bot applies a Landlock LSM sandbox after startup. System paths are read-only. Bot storage, config, cogs, Reticulum config, and temp directories stay writable.

bot = LXMFBot(
    name="SecureBot",
    landlock_enabled=True,
)

status = bot.get_landlock_status()
print(status)

Environment overrides:

• LXMFY_LANDLOCK=0: disable Landlock
• LXMFY_LANDLOCK=1: force an attempt on Linux
• unset: follow landlock_enabled and kernel auto-detection

External script cog sandbox

Executable cogs in cogs/ can run in a restricted environment when external_cogs_sandbox_enabled=True (default). Set external_cogs_sandbox_type to:

• auto (default): prefer Landlock, then bwrap, then firejail
• landlock: Landlock-only via preexec_fn
• bwrap: bubblewrap read-only bind sandbox
• firejail: firejail private profile with no network
• none: no subprocess sandbox

See the docs for full configuration details.

Development

• Python 3.11+
• Poetry

poetry install
poetry run lxmfy run echo

Common Makefile targets:

make lint       # ruff check
make typecheck  # pyright lxmfy
make test       # pytest
make ci         # lint, typecheck, security check, test, build

Contributing

Send ideas and issues to LXMF: 7cc8d66b4f6a0e0e49d34af7f6077b5a

License

0BSD