## Best for
- +General-purpose assistant runtime with broad tool access and channel integration
- +Iterative, operator-driven workflows over strict vertical specialization
- +Teams who can enforce security discipline around tool/skill boundaries
## Avoid when
- +Strict enterprise policy/compliance envelope (consider cloud-native stacks)
- +Graph-heavy deterministic orchestration with explicit state transitions
- +Low-code visual workflow ownership by non-engineering teams
## Deployment
Local/devSelf-hosted daemonHybrid (local + gateway)
## Security & reliability
Plugin/skill supply-chain risk; local privilege escalation if not locked down. No built-in sandbox; relies on operator discipline and optional containment.
- ·Long-running sessions with context retention
- ·Multi-channel delivery (retries vary by channel)
- ·No built-in checkpointing or durable replay
## Integrations
Broad channel adapters (WhatsApp, Telegram, Slack, Discord, etc.). Model-agnostic (OpenAI, Anthropic, OpenRouter, local). Skill/plugin ecosystem; MCP not native.
## Cost
Free/open source. Infra cost = your hosting + model API usage. Predictable at small scale; variable with channel and model choice.
## Supported Models
ClaudeGPT-4oDeepSeekLocal LLMs
## Chat Channels
WhatsAppTelegramSlackDiscordiMessageSignal
## System Requirements
Node.js 18+, 4 GB RAM recommended