Turnstone is a Mac desktop platform that runs a coordinated team of AI agents powered by external model accounts, API keys, or local inference engines, keeping agent memory and connected workplace data under local control.
What the product does and how it is positioned
Turnstone is an agentic workspace designed for macOS that coordinates specialized artificial intelligence personas to handle ongoing administrative, research, drafting, and operational tasks. Instead of requiring proprietary token meters, the application allows users to connect their existing subscriptions from providers like OpenAI, Anthropic, xAI, and Cursor, enter developer API keys, or run open models locally via Ollama.
The environment centers on a Chief of Staff agent that distributes work across dedicated sub-agents tailored to particular business functions. Turnstone periodically ingests data from external productivity tools such as Gmail, Slack, Granola, and Google Calendar, updating agent context and drafting communications that reflect the user's personal writing patterns while keeping final sending authorization in human hands.
Source-supported ways to use the product
Configuring an automated morning routine where an agent compiles scheduled calendar meetings, pending replies, and urgent obligations into a consolidated briefing.
Generating contextual email responses and beta invitations that mirror individual writing patterns, holding each message for manual approval prior to dispatch.
Directing a Chief of Staff agent to delegate launch post drafting to a specialized writing agent, switching underlying language models dynamically during the session.
The documented workflow, where available
Define initial operational priorities with the primary coordinator agent, which establishes specialized sub-agents for specific project roles.
Sign in with existing AI provider subscriptions, local Ollama environments, and productivity applications like Gmail, Slack, and Google Calendar.
Set recurring reporting cadences and authorize agents to browse web pages, research topics, or fill online forms in dedicated browser tabs.
Inspect generated summaries, code modifications, and draft communications directly within the interface before approving external transmission.
Turnstone structures autonomous assistance around role-specific agents managed by a central Chief of Staff. Rather than binding all tasks to a single general-purpose prompt, users assign distinct responsibilities—such as engineering analysis, outreach drafting, or launch copywriting—to separate agents. These agents can transfer context and work products between one another, returning finalized summaries to the main user interface.
A core element of the architecture is model independence. Users can map individual agents to different foundation models based on task suitability, utilizing accounts from OpenAI, Anthropic, xAI, or Cursor, as well as API connections to Gemini or local execution via Ollama. Turnstone preserves conversation state, learned behavioral rules, and persistent memories even when switching models mid-thread.
Checks to run with your own material and workflow
What was checked and when
Answers based on the source-checked product record
All agent configurations, chat logs, and operational memories are stored locally on the user's Mac in an accessible file folder.
Users can connect accounts for ChatGPT, Claude, Grok, and Cursor, provide API keys for models such as Gemini, or execute local models using Ollama.
Yes, users can switch language models in the middle of a conversation while the agent retains its designated role, context, and accumulated memory.
No, data gathered from connected workplace applications is used strictly to build the individual user's agent memory and is not used for model training.
Agents prepare communication drafts based on learned writing styles, but messages remain in draft form until the user manually reviews and sends them.
