# One Twin, Many Interfaces > Use the same reviewed business context in the Twin workspace, ChatGPT, Claude or your terminal. The interface changes; the access and evidence boundaries stay. Published: 2026-07-23T08:00:00Z Updated: 2026-09-10 Author: Archie Norman (Founder, MLX) Category: Twin Tags: twin, interfaces, sdk, governance, mcp, agents, evidence Canonical URL: https://mlx.systems/blog/one-twin-many-interfaces ## TL;DR - The workspace, enabled AI connections and Twin CLI reach shared business context through governed interfaces. - Each route preserves caller identity, organisation scope, permitted access and the answer's evidence requirements. - A documented integration still needs to be enabled for your organisation and used through protected sign-in. --- A finance lead asks a question in the Twin workspace. A colleague asks through Claude. An engineer uses a terminal. Each wants the same business meaning: the definition the team has reviewed, with the evidence needed to understand the answer.
Figure 01. Four illustrative presentations of one question. The reviewed definition and access boundaries stay with the Twin.
## Start With the Business Meaning The useful thing to share is not a prompt or a chat transcript. It is the model of the business underneath: what a customer is, which invoices count, how a metric is defined and which evidence supports it. That meaning belongs to the Twin. An interface gives a person or agent a way to use it. If every new client has to reconstruct the definition from scratch, the company has gained another interpretation to reconcile rather than another useful way to work. ## Choose a Door That Fits the Work The Twin workspace brings sessions, channels, files and views together. It is the natural place to explore connected context, inspect results and continue work with a team. ChatGPT and Claude provide protected connection routes for organisations that enable them. Codex and Claude Code can use a user-scoped remote Twin connection through the organisation's connection pack. The Twin CLI provides a terminal route, while managed implementations can use the public REST and TypeScript SDK surfaces that are enabled for them. These are documented routes, not a claim that every organisation already has every one configured. [Connection availability](https://docs.mlx.systems/docs/connect/availability) is the maintained reference; [choose where to work](https://docs.mlx.systems/docs/connect/choose) explains the options. ## What Crosses Every Boundary An interface has to preserve more than the question. The signed-in caller determines authority. The organisation determines scope. Capability policy limits what the request may do. The permitted context and reviewed definition determine what it may rely on. The answer's evidence must survive that journey too. The reader needs to know what supports the result and where the proof ends. A different chat layout must not turn a narrow evidence claim into a broader one. This does not promise byte-for-byte identical answers across clients. Wording, visual presentation and supported features differ. The shared contract is the business context and its access and evidence boundaries, not a fixed paragraph of prose. ## Public Software, Private Access Anyone can install the Twin CLI. That does not let them query an organisation's data. Browser sign-in establishes who they are and which organisation and channels they can access. An organisation enabling an AI connection is also distinct from a user signing into it. Both the route and the caller's authority matter. A client cannot give itself wider access by placing a different organisation name in the question.
A closer look at the terminal route The public CLI reaches the shared HTTPS gateway. After installation, `twin login` opens the protected browser flow; the CLI stores the resulting credential in the operating system's credential store. A natural-language query uses the documented form: ```bash twin query --question "What changed this week?" ``` The CLI also exposes sessions, channels, artifacts and feedback. Use its help and the [current CLI guide](https://docs.mlx.systems/docs/connect/cli) for supported commands and access requirements.
## A Model Provider Is a Different Choice Where someone asks a question is separate from which model performs a configured task. An organisation can have a model provider configured without having a self-service connection from that provider's consumer application. For example, a configured Gemini model route does not establish that the Gemini consumer product has an enabled Twin connector. The [availability guide](https://docs.mlx.systems/docs/connect/availability) makes this distinction explicit. Start from the route your organisation has enabled, then check its access and data controls. ## The Contract Outlives the Interface Chat clients and protocols will change. The reviewed business context should remain useful through those changes, along with the controls that govern who can use it. That is the payoff of building the Twin once: another way to ask does not have to become another definition of the business. To understand the evidence that should accompany an answer, continue with [Every AI Answer Should Come With a Receipt](/blog/every-ai-answer-needs-a-receipt).