I build local and hybrid AI systems, bounded agent workflows, and source-first knowledge architectures.
This profile repository is the public entry surface for my ecosystem. It helps humans and agents find the right repository, but it is not the source of truth for the specialized layers themselves.
Two long-horizon directions anchor most of the work:
- Agents of Abyss (AoA): an operational federation of explicit layers for long-horizon agentic systems
- Tree of Sophia (ToS): a source-first living knowledge architecture for philosophy and world thought
Beneath them sits abyss-stack, the infrastructure substrate. Alongside them sits ATM10-Agent, a local-first companion surface shaped by the same broader direction.
For a compact ecosystem vocabulary, see GLOSSARY.md.
- for the ecosystem center, constitutional map, and federation rules, start with Agents-of-Abyss
- for the knowledge world and source-first architecture of thought, start with Tree-of-Sophia
- for the runtime body beneath AoA and ToS, go to abyss-stack
- for a local-first companion application surface, go to ATM10-Agent
- for reusable practice, execution, and proof, move through aoa-techniques -> aoa-skills -> aoa-evals
This repository is a coordination and orientation surface.
It exists to:
- introduce the broader AoA / ToS ecosystem
- point to the repositories that own the real charters, roadmaps, workflows, and implementations
- keep the public layer map legible for humans, smaller models, and new collaborators
It should not silently replace the source repositories it links to.
-
Agents-of-Abyss
Constitutional center of AoA: ecosystem identity, layer map, federation rules, and program-level direction -
Tree-of-Sophia
Living knowledge architecture for philosophy and world thought -
abyss-stack
Infrastructure substrate for AoA and ToS: modular, rootless, local-first runtime, deployment, storage, and lifecycle services for self-hosted AI systems -
ATM10-Agent
Local-first ATM10 companion for Windows 11: vision, RAG/KAG memory, dry-run automation, voice, and a gateway-backed operator panel
-
aoa-techniques
Public canon of reusable engineering techniques for coding agents and humans -
aoa-skills
Bounded agent-facing skill bundles composed from reusable techniques -
aoa-evals
Portable evaluation bundles for agents and agent-shaped workflows -
aoa-routing
Thin routing layer for explicit navigation, typing, and dispatch across AoA surfaces -
aoa-memo
Memory and recall layer for explicit, reviewable, provenance-aware memory surfaces -
aoa-agents
Role and persona layer for explicit agent profiles, role contracts, and handoff posture -
aoa-playbooks
Scenario and composition layer for recurring operations, handoffs, fallback paths, and validation posture -
aoa-kag
Knowledge substrate layer for provenance-aware derived structures, graph-ready projections, and framework-neutral retrieval surfaces built from source-first repositories
I am building toward:
- agent systems that stay legible as they scale
- knowledge architectures that accumulate layers without collapsing into noise
- infrastructure where new tools become durable capabilities
- local-first and hybrid systems that keep human meaning and operational clarity in view
- publish durable techniques, not one-off accidents
- keep source-of-truth boundaries explicit
- prefer modular growth over brittle fusion
- make workflows reviewable, reproducible, and portable
- keep systems legible to humans while accelerating agents
- humans and agents in the loop, with clear handoffs
- compact public surfaces over inflated repository folklore
- layered growth that does not erase ownership boundaries
- Systems: Fedora, Podman, Windows 11
- Languages: Python, Bash, JavaScript, PowerShell
- Agent & app layer: FastAPI, Uvicorn, Streamlit, LangChain, LiteLLM, n8n
- Models & inference: OpenVINO, OpenVINO GenAI, Ollama, Transformers, Torch
- Data & memory: Postgres, Redis, Neo4j, Qdrant
- Observability: Grafana, Prometheus, Alertmanager
- Build workflow: ChatGPT, Codex, GitHub