Allegedly, actually aimless.
Aimlessly is a directed (non-acyclic) decision graph executor — the procedural memory layer for autonomous agents.
Part of the Christmas Island agent stack:
| Library | Role | Status |
|---|---|---|
| Allegedly | Knowledge graph (declarative memory) | Implemented (v6) |
| Actually | Meilisearch wrapper (queryable knowledge surface) | Designed |
| Aimlessly | Decision graph executor (procedural memory) | Design phase |
| Apparition | UI for all of the above | Named |
Synthetic tools as decision opcodes. Each step in a workflow is a micro-prompt with fake tools whose names force a deterministic, structured decision. The LLM is the transition function; the tools are the opcodes; the graph is the program.
Key properties:
- Fresh synthetic conversations per hop — no context drift, no hallucination compounding
- Non-acyclic by design — cycles are features, not bugs, with meta-inspector oversight
- AI-editable graphs — workflows learn and improve themselves
- Inter-workflow composition — graphs call each other like function calls with a call stack
- Model-agnostic routing — cheap local models for simple decisions, escalate to heavy cloud models when needed
Early design phase. See docs/ for design documents and assessments.
MIT