It's a claw that has been lobotomised. All the useful framework that these *claw-likes do but without the infinite security nightmare.
it's a scalable, resilient, SECURE AI agent that communicates over discord (or others if you want) and does things for you, whilst not being weird about it.
It is very cute, says things like
Lobber not like BEEEEEEP. Lobber say hello! Lobber have add tool. Need more tool? Lobber can get. Lobber try add text tool. Text tool not exist. Lobber not know what to do.
we love lobber
- Ability to draft its own tools for later use (manual review required)
- Hot reloading of custom tools and system prompt via
!!reload - Persistent memories and identity management
- Lazy tool loading, so JSON schemas don't eat your context om nom nom
- the agent uses a meta-tool
add-toolwhich provides the tool schema to it on the next turn
- the agent uses a meta-tool
- Retained chain of thought so your lobber can remember what it was thinking when it did something
- Lobber's own little cave where it stores shiny rocks (markdown files) for later use
- NO SILLY BASH EXECUTION WE'RE ALL ELIXIR ALL THE WAY DOWN
- Dynamic context compaction via
!!compactto keep your conversations going - Cool routing system to send to different providers/models based on the task type
- Json-backed crontab to run stuff when you want to I guess
graph TD
subgraph Channels
CSV[Channels Supervisor] --> |spawns|DSV[Discord Supervisor]
DSV --> |spawns|DWS[Discord Websocket]
CSV[Channels Supervisor] --> |spawns|O[Other Channel]
end
subgraph Conversations
DWS -->|get_or_spawn| CONVS[Conversations Supervisor]
CONVS -->|spawns| C1[Conversation discord:123]
CONVS -->|spawns| C2[Conversation discord:456]
DWS --> |add_message| C1
C1 --> |send_response| DWS
end
subgraph Agent
ASV[Agent Supervisor]
ASV --> |spawns|AT[Agent Task]
AT --> C1
C1 --> |prompt| ASV
end
subgraph Provider
DL[Delegation]
OR[OpenRouter]
DL --> OR
DL --> XM[Xiaomi MiMo]
end
AT --> DL