| ⬡ | Input | I/O | Entry point — receives the user message and chat history | system_prompt |
| ⬡ | Output | I/O | Final response — text shown in chat or returned to caller | — |
| ⊕ | Split | Flow | Duplicate input to multiple outputs for parallel processing | — |
| ⊗ | Merge | Flow | Combine outputs from parallel branches (concatenate or pick first) | mode |
| ⑂ | Condition | Flow | Branch on keyword contains / regex / length. True port if match, False port otherwise | mode, value |
| ✎ | Injector | Flow | Prepend a fixed prefix to the message (system instructions, personas) | prefix |
| ⟨⟩ | Formatter | Flow | Transform output: uppercase, lowercase, trim, JSON extract, strip markdown | mode |
| ⚡ | Groq | LLM | Groq cloud inference — fastest option. Llama 3.3 70B by default | model, temperature, max_tokens |
| 🦙 | Ollama | LLM | Local LLM on your Mac via Tailscale — fully private, no data leaves network | model, temperature, max_tokens |
| ↗ | OpenRouter | LLM | Multi-provider routing — access 100+ models via one API | model, temperature, max_tokens |
| ♊ | Gemini | LLM | Google Gemini Flash — fast and cheap for summarization | model, temperature, max_tokens |
| 〽 | Mistral | LLM | Mistral AI models — strong for European data / multilingual | model, temperature, max_tokens |
| ⬡ | Cerebras | LLM | Cerebras WSE — ultra-fast inference for latency-sensitive tasks | model, temperature, max_tokens |
| ☁ | Cloudflare AI | LLM | Cloudflare Workers AI — serverless edge inference | model, max_tokens |
| ⬡ | SmartRouter | LLM | Your SmartRouter service — auto-selects best model based on query type | temperature, max_tokens |
| 🔍 | Brave Search | Tools | Web search via Brave API — returns top results as context | num_results |
| 🦆 | DuckDuckGo | Tools | Free web search via DuckDuckGo — no API key required | num_results |
| 📄 | Jina Reader | Tools | Fetch and extract clean text from any URL | — |
| 🎨 | Image Gen | Tools | Generate images from text prompts via Cloudflare AI | — |
| $ | Bash | Tools | Execute shell commands on the Linode server (restricted — dangerous commands blocked) | command |
| { } | Codebase | Tools | Search your codebase: grep for patterns, read specific files, find files | operation, path |
| 🧠 | RAG Memory | Memory | Retrieve relevant context from Supabase pgvector — your conversation history | top_k |