Open Source

Local-first MCP tooling for controlled Codex CLI automation. codex-mcp-wrapper is a local-first MCP server around Codex CLI. It is built for maintainers who want AI-assisted code work to be scoped, inspectable, and reviewable through MCP-compatible clients. It supports read-only planning jobs, isolated apply jobs, job status and result APIs, diff inspection, review and resume flows, allowed repository roots, conservative sandbox defaults, and explicit safety gates. This is the strongest public bridge between inAi's Open Source portfolio and Products for Agents: software that agents and AI-assisted workflows can call, inspect, and use with clearer boundaries.

A host-first operating-surface prototype for wearable AI. PrismOS is an experimental operating-surface project for Brilliant Labs Monocle hardware. The host device owns app state, layout, provider calls, persistence, diagnostics, recovery, and bounded display output, while the Monocle stays closer to a thin display, input, and media endpoint. It explores compact AI-assisted apps and interface patterns for hardware-adjacent workflows, including concepts such as Oracle, Vision, Signal, Chronos, Teleprompter, Babel, and Gallery-style experiences. PrismOS shows that AI-native products do not have to live only in web dashboards or chat windows.

A desktop workspace for large image-generation runs. ImageAI Batch Generator is a desktop workspace for managing larger image-generation workflows. It supports prompt imports, shared settings, per-row overrides, queueing, throttling, retries, cancellation, autosave, export metadata, and offline mock generation. It is designed for the moment when image generation moves beyond one prompt at a time and becomes a repeatable workflow that needs structure, tracking, retries, and operational control.

A local tool for live transcription, translation, and conversation assistance. WhisperMind is an experimental local app for live microphone transcription, translation, and conversation assistance. It includes desktop and web-interface variants around speech, translation, and realtime conversation workflows. The project explores how AI can support live communication: not only by answering prompts, but by helping people understand, translate, and respond in the flow of conversation.

An older prototype for building and running multi-step AI workflows. ChainerAI is an older prototype for composing multi-step AI pipelines. It includes a backend, frontend, saved chains, ordered prompt/model steps, persistence, login, containerized setup, and provider-adapter-style workflow ideas. It is not presented as a current flagship product. It remains public as a legacy prototype and reference point for AI workflow composition.

A lightweight local browser UI for one configured assistant. ChatGPT Assistants UI is a small local Flask interface for chatting with one configured Assistants API assistant. It stores local chat metadata and provides a simple browser-based interface for creating, renaming, selecting, and deleting chats. It is preserved as a legacy reference project: simple, focused, and useful mainly as an example of an earlier local assistant interface.

Some Open Source work, especially codex-mcp-wrapper, connects directly to the agent era: tools and interfaces designed for AI-assisted workflows to call and operate.

AI should be understandable beyond technical circles. AI for Everybody explains AI, agents, risk, work, and public questions in accessible language.