AI-Asassin
A general-purpose AI assistant that condenses your dialog into a graph of statements and answers every reply from that graph — never from a transcript.
The AI revolution is starting — right now
In the short term, everything that can be done on a computer will be done through AI or with AI assistance. The gap between what AI can already do and what trading tools provide is widening every month.
What AI-Asassin does
A conversation with AI-Asassin starts as an ordinary AI chat. Iteratively, the dialog is transformed into a directed acyclic graph (DAG) of statements — the Synergy Context Gate (syngate): a sentence that survived the exchange becomes a node, a short statement with the items it follows from as its parents, and the chat goes on anchored at items rather than at the tail of a transcript. The graph is then the context engine: a reply anchored at an item is answered from exactly the reasoning that leads to the point under discussion and what details it — not the whole history, and not a lossy summary of it. Each reply either continues the dialog or condenses it back into the graph, and you keep the last word by reviewing what the graph asserts.
Items are plain text files in your repository — a YAML front matter over a markdown statement — shaped only by their parents. The AI agent restructures the graph through tree calls — add, remove, move — never by editing files, so every change is explicit and attributable.
The seed context of an item is the chain of its ancestors' statements, root first, then its own, then those of all its descendants — every item once, in order. That selection, and only that selection, is what a reply anchored at the item is answered from, so the model receives neither too little nor too much.
Two optional disciplines turn the graph into a gate. Review stamps an item as approved by you, so the graph cannot drift from what was agreed. Test binds an item to executable routines, frozen by hash on approval — test-first, while still red — so the graph cannot drift from what the code does.
The local editor shows the graph as an outline and as one structured document, with the AI chat inside every item's panel; the CLI is the same tree calls from the terminal. AI models are reached through connectors — the model, effort and call modes are chosen per exchange.
How Open Trader is built with it
Product architecture definition through dialog analysis and synthesis is one use case of this loop. With test and review on, the graph of a software project is its architecture and contract tree: every leaf definition is bound to a test frozen on approval, and an item's seed context is the context an AI coding session is seeded from. Open Trader is such a project — its CI runs the gate on every push — and so is AI-Asassin itself.
pipx install git+https://github.com/l2xl/ai-scratcher.git,
then run syngate ui inside a managed project.