An agentic AI engine for software development — API-first, CLI on top.
The engine behind everything I build — a proprietary agentic layer over a frontier language model that powers my own systems through a REST / WebSocket API and a published CLI client. Built for multi-step reasoning with real tool execution, not just chat completion.
Multi-step reasoning with tool and action execution — the engine plans, calls tools, and acts on results.
REST + WebSocket API, plus a Model Context Protocol (MCP) server for device and tooling integration.
pgvector + fastembed embeddings for fast, meaning-aware cache hits.
Redis-backed sessions keep conversational state durable across requests.
@exybit/cli — a TUI chat client on npm that talks to the engine over WebSocket.
The first-person AI persona on exybit.org — the portfolio assistant — is powered by this engine.
Terminal chat client, npm i -g @exybit/cli, published and live.
An autonomous R&D pipeline whose workers (fission-worker, on npm) execute tasks in isolated Docker containers.
Powers the AI chat inside the Lotus Commercial operations portal, answering inquiries across jobs, scheduling, attachments, invoices, and more.
A conversational personal assistant (Telegram + web) that tracks todos, reminders, expenses with receipt-photo capture, and notes — surfacing daily digests and spending insights, and extending itself with new skills on demand.