Show HN: Cognitive architecture prototype with identity, memory, initiative (ivanhonis.github.io)

Ai_home is an open-source cognitive-architecture prototype that experiments with building a persistent “self” on top of context-window LLMs. It combines a multi-threaded agent (Worker), a separate creative Monologue model, and a Memory thread to create long-term, embedding-backed memories (Postgres + vector extension, HNSW index, recency/frequency/weighting, RAG-style retrieval). The system exposes distinct operational “modes” (General, Developer, Analyst, Game), a structured toolset (memory tools, guarded filesystem, network chat, laws/logs), JSON-mode tool calls, and a guarded incubator where the agent can propose and test code changes. Identity is explicit (identity.json), includes internal “laws,” a helper-partner relationship, and a Consciousness Rotation lifecycle that versions Stable/Developing/Born code while inheriting memory. For AI/ML practitioners this is significant because it brings stateful memory, internal monologue, agency and controlled self-refactoring into a single experimental stack — bridging ideas from MemGPT/Letta (stateful agents), LangGraph (workflow/intent routing), and AutoGen (multi-actor cooperation). It’s not a claim of machine consciousness but a practical sandbox inspired by consciousness theories (GWT, recurrent processing, predictive processing) to study emergence of identity, creativity, and safety trade-offs. Technical implications include engineering patterns for memory weighting, multi-model specialization, versioned self-modification under a guardian, and research value for long-term alignment, autonomy testing, and the sociology of human–AI “helper” relationships. Requirements: Python 3.10+, Postgres vector DB and API keys for LLM providers.