Autonomous Cognitive Architecture — LLMs as tools, not agents
Archotec is a cognitive agent architecture where the LLM is a perception encoder and language renderer — not the decision maker. Decisions emerge from a belief-driven stochastic policy that runs entirely outside the model.
Not a chatbot wrapper. Not a prompt orchestrator. Not a ReAct loop.
| Chatbot Frameworks | Archotec | |
|---|---|---|
| Decision mechanism | LLM output routing | Stochastic policy (EFE) |
| State representation | Conversation history | Continuous belief distribution |
| Action selection | argmax / intent match | Softmax sampling |
| LLM role | Controller | Perception encoder + renderer |
Exposes the SDK interface contracts: the typed protocols every module must implement, and the data types that cross module boundaries.
from archotec_sdk.protocols import PerceptionModule, PolicyModule
from archotec_sdk.types import Observation, BeliefSnapshot, CycleContextSee sdk/protocols.py and sdk/types.py.
Apache 2.0 — see LICENSE