# Your Twin

> Your Twin connects business records, relationships and rules into a shared understanding for your team and AI tools. Ask questions and check the sources.

## What your Twin offers

Your Twin connects business records, relationships and rules into a shared understanding for your team and approved AI tools. Mercury Labs builds the software and offers engineering support for setup and integrations.

A customer receives:

- **A managed deployment** with agreed data access and AI providers: a shared deployment on Team, or a dedicated, managed server on Core, Scale and Enterprise.
- **Connections to your existing systems**, starting read-only. Keep managing the original records where they are.
- **Connected business knowledge**: records, relationships and agreed definitions, such as what counts as revenue.
- **Checked versions you can review**: calculations are tested against your data and the design is independently evaluated before release. It stays off until you switch it on.
- **Figures calculated by a graph engine**: the engine holds your connected records and agreed definitions in memory and works out each figure in milliseconds. It uses no AI model for the arithmetic, so the same question gives the same number wherever it is asked, each figure traces to the records behind it, and it reads only the records the person asking may see.
- **Answers with sources and checks** when using the checked Twin. This is called the governed answer path; questions it cannot support are refused.
- **Model choice and cost control**: use the configured AI tools that fit the task without rebuilding your Twin.
- **Optional engineering support** from Mercury Labs for source connections and the first workflow.

## The problem your Twin solves

Understanding the work should not mean chasing updates, rebuilding spreadsheets and asking which number is right. Your Twin connects customers, projects, finances and commitments into one shared picture, with definitions and sources your team can inspect.

Your existing systems remain the source of truth. The Twin keeps the agreed business definitions, evidence and version used for a governed answer. Changing the AI provider does not require rebuilding that knowledge.

## How it works

The landing page tells it in four steps: the records scattered across your systems, one customer followed across them, the shared understanding your Twin builds, and the same answer in the tools your team uses.

1. **Connect the records.** Software checks relationships across the systems you choose. Sensitive values are converted to keyed fingerprints before comparison; the authoring AI uses descriptions and measurements, not raw records.
2. **Build and check.** An AI proposes the design. Software generates the code and tests. Calculations must run and reconcile against source data. A separate AI evaluates the work and cannot repair or approve the design it reviews. Failed checks block release.
3. **Review before switching on.** Each checked version is published as a sealed release with its test results and source references. It stays off until activated. Source freshness is shown where the configured product and interface expose it.
4. **Put it to work.** Your team can use the Twin through approved AI tools. Agents can use the checked Twin or explore connected records within their permissions. Only governed answers identify the published definition and version; exploratory work is kept distinct and needs review. Agents can save results as structured records in shared channels through permitted write operations.
5. **Keep it current.** New evidence can propose changes. Important changes are tested and published as new versions for review.

## Keep the understanding useful

Connected records refresh on the schedule agreed for each source. Changes to business definitions are checked and released for review, with version history. Your Twin keeps this understanding available across approved AI tools. The subscription provides the managed deployment and plan support; setup, additional connections and engineering work are agreed separately.

## Keep the work in context

Channels bring related records, saved views, files and conversations together around areas such as Commercials, Delivery or Finance. Teams can work through a review or follow-up without losing the supporting records. Inbox brings incoming updates and suggested actions into a review queue; it is not a promise to detect every business exception automatically. Access rules still apply.

## Trust and deployment boundaries

- On Core, Scale and Enterprise, your Twin runs on a dedicated server. Current standard hosting is in EU or US regions; UK deployment is in development. Team runs on a shared deployment.
- Connected records are stored within the configured deployment for your organisation.
- Current configured model routes use external providers. Agree which provider receives each request and what business information is sent before setup. Server geography does not determine model-processing geography.
- The standard governed query path is limited to your organisation and is read-only. It does not include browser, terminal or write access.
- Shared and production changes require verified authority and explicit review.
- Governed answers identify the published definition and version used, with source references. Freshness is shown where the configured product and interface expose it. Individual rows can be traced to their source only where the published definition preserves that link.
- Model changes require the relevant testing and controlled rollout.

## Who your Twin is for

Established organisations whose teams prepare recurring reviews and decisions from information spread across several systems. Start with a specific question, the records behind it and the people who will check the answer.

### Accountancy practices

For managing partners, practice managers and client-service teams. Connect Xero, SunSystems and Slack alongside Google Workspace and Microsoft 365. See which clients need attention, what is missing and who is following up. Agent results can become client trackers and review tables in shared channels. Your team makes the professional judgements and signs off accounts.

- [Your Twin for accountants](https://mlx.systems/for/accountants): Client reviews, missing evidence and follow-up owners.

### Consultancies

For consulting partners, operations directors and delivery leads. Connect plans, documents and conversations across Google Workspace and Microsoft 365. See what changed, where scope is growing and which client decisions are holding up delivery. Agent results can become decision logs and project trackers in shared channels. Your team approves client messages, scope changes and delivery commitments.

- [Your Twin for consultancies](https://mlx.systems/for/consultancies): Delivery reviews, client decisions and handovers.

### Your Twin, through your tools

Use Twin chat or connect approved external tools through MCP, the Twin CLI, SDK or API. The site illustrates ChatGPT, Claude, Gemini and Microsoft 365 Copilot; in Microsoft 365 Copilot and Teams, your Twin appears as an agent your team can ask. Each connection requires client setup, permissions and deployment configuration. The records and interfaces shown are examples, not evidence of a live customer integration.

## Project Venus research

Project Venus is our open weight model research initiative, which gives you sovereignty over your entire AI stack. It serves from dedicated infrastructure pinned to the UK or EU, while researching learning business terminology and decisions with current facts kept in the Twin, sources and limitations staying visible, and the measured speed and cost of operating a private model. Enquiries can be made through the research page form.

## Pricing

Two ways to buy: Team, priced per user on a shared deployment, and three plans that each have a dedicated, managed deployment. Full detail at https://mlx.systems/pricing.

- **Team**: £75 per user per month, on a shared deployment. A team that works in cloud apps: cloud app connectors, daily refresh, up to 12 months of history per source, a usage allowance included and self-serve standard support.
- **Core**: from £3,000 per month. One focused team: an initial source set, daily refresh, up to 12 months of history per source, up to 10 users and standard support.
- **Scale**: from £5,000 per month. More teams, workflows and sources, higher-frequency refresh, unlimited users and a dedicated support channel.
- **Enterprise**: from £15,000 per month. Tailored controls, environments and support for complex organisations, custom model routes and a named service lead.

AI usage: Team includes a usage allowance, with top-ups available. On Core, Scale and Enterprise, model credit packs are £250 (25,000 credits), £750 (100,000 credits) or £1,800 (300,000 credits) per month.

Optional engineering: Mercury Labs engineers help with setup and integrations at £1,200 per day. We agree the work for your teams and systems, model rates and any unusual hosting requirements before starting.

Start with the question your team keeps chasing. A focused, capped two-week activation connects the systems for one workflow, builds and checks your Twin, then lets your team review it against real examples before switching it on. Setup, ongoing support and changes are agreed with you. Contact team@mercurylabs.io to discuss your first workflow.

## Common questions

### Where does our data go?

Your Twin reads the systems you connect without changing the original records. On Core, Scale and Enterprise it runs on a dedicated server for your organisation; Team runs on a shared deployment. Keep managing the original records in your existing systems.

When you use an external AI provider, the business information needed for that request is sent to the provider configured for your organisation. We agree the connections, access and model provider before setup.

### Do we need to migrate our legacy systems first?

Keep your existing systems. We agree a first workflow and connect the systems it needs, starting with read-only access.

Mercury Labs engineers work with your team to connect the relevant records, agree business definitions and test the calculations against your data. You review the result before switching it on.

### Can we choose models and control their cost?

Yes. Use stronger models for demanding work and lower-cost models for routine tasks. Your Twin keeps the same business records and definitions when you change AI providers.

Each provider needs to be configured for your organisation. Important changes are tested before they go live.

### What does this cost?

Subscription: Team is £75 a user a month, on a shared deployment. Core starts at £3,000 a month, Scale at £5,000 and Enterprise at £15,000, and each of those includes a dedicated, managed deployment.

AI usage: Team includes a usage allowance, with top-ups available. On Core, Scale and Enterprise, model credit packs start at £250 a month.

Optional engineering: Mercury Labs engineers can help with setup and integrations at £1,200 a day.

See the pricing page for what each plan includes. We agree the work and costs for your teams and systems before starting.

### Why not just use ChatGPT or Copilot?

Keep using them, with your Twin connected. On their own they search your systems one at a time, so an answer depends on which system was searched and how the question was worded. Your Twin joins your business records and agreed definitions into one picture, so ChatGPT, Copilot and your team get the same answer, with the sources and calculations it used.

Keep your business knowledge in the Twin as your choice of AI tools changes. Each external connection needs the right setup and permissions.

### How does the Twin work out its numbers?

Its graph engine does the arithmetic. The engine holds your connected records and agreed definitions in memory and works out each figure in milliseconds. It uses no AI model, so the same question gives the same number wherever it is asked, until the records change.

The AI model plans the question and writes the answer round the figure. Each figure comes with what was counted and how, and the engine reads only the records the person asking is allowed to see.

### How does our Twin stay useful over time?

Connected records refresh on the schedule agreed for each source. When a business definition changes, it is checked and released for review, with version history. Your team and approved AI tools keep using the same shared understanding.

Your subscription covers the managed deployment and support included in your plan. We agree setup, additional connections and engineering work with you. Expanding the Twin starts with the next useful question, rather than rebuilding it for each AI tool.

### Where do we work through the results?

Channels bring related records, saved views, files and conversations together. Organise them around areas such as Commercials, Delivery or Finance, then work through a review or follow-up with the supporting records close by.

Your Inbox is where you review incoming updates and suggested actions, with the context behind them. Start an agent session from an item when it needs more work. What you can see and do depends on your access.

### We're regulated. Can we actually use this?

We need to check your requirements with you. Your Twin starts with read-only connections and access limited to your organisation. You can review each tested version of the Twin before it is switched on.

The standard checked query path does not include browser, terminal or write access. Your team still needs to approve where data and model requests go, how long evidence is kept and how the deployment is operated.

### What do the AI agents actually see?

While building the Twin: the AI agents use descriptions and measurements, not raw records. Software converts sensitive values into protected fingerprints before comparing them to find connections.

When you use it: agents can use the checked Twin or explore connected records within your organisation’s permissions. Exploratory work is kept distinct from answers produced through the checked Twin.

### Which systems can you connect?

Accounting and payments: Xero, SunSystems, Stripe and GoCardless.

Documents, email and collaboration: Microsoft 365, SharePoint, Outlook, Teams, Google Drive, Docs, Sheets, Gmail, Google Calendar, Slack, Granola and Notion.

CRM and delivery: HubSpot, Linear and GitHub.

You can keep legacy ledgers such as SunSystems. We confirm which records and connections your first workflow needs before setup.

### What happens when the Twin cannot support a question?

It tells you when the checked Twin cannot answer. This path (called a governed answer) identifies the published definition, version and source freshness it used. If the question or evidence is unsupported, it returns “not supported”.

An agent can still explore the underlying records within its permissions. That exploratory work is not labelled a governed answer and needs to be reviewed accordingly.

## Customer evidence

- [LVF Creative Studio](https://mlx.systems/case-studies/lvf-creative-studio) (4 hrs reclaimed on one counts task): LVF brought project delivery, finance, floral counts and team actions into one AI-ready workspace, reclaiming four hours on a single counts task and giving the team a clearer way to run complex events.

## Writing

Most recent first. Each post has a corresponding markdown rendering at `/blog/{slug}/markdown`; the full corpus is at `/llms-full.txt`.

- [What a Decision Model Taught Our Agent Harness](https://mlx.systems/blog/what-a-decision-model-taught-our-agent-harness): Jev chose the data for an answer in a few hundred milliseconds and missed only one kind of question: a near-tie our code read as a firm no. Fixing the same habit in our own harness raised offline coverage from 58% to 82%.
- [How a Digital Twin Is the Key to Running a Portfolio of Fast-Growing Companies](https://mlx.systems/blog/running-a-portfolio-with-a-digital-twin): How I run four fast-growing companies from one seat, what a digital twin actually is, and why playbooks - not the model - turned out to be the unlock.
- [Three Cents of Probing Says Jev Is a Small Qwen in a Very Good Suit](https://mlx.systems/blog/fingerprinting-jev-for-three-cents): I sent Jev about 120 requests and roughly $0.03 of API spend. The token counts, the 32,768 context limit and the punctuation behaviour say pretrained Qwen-class LLM, customised tokenizer and parallel scoring heads. No new science, but genuinely good engineering.
- [What Should a Twin Learn From a Mistake?](https://mlx.systems/blog/what-should-a-twin-learn-from-a-mistake): An overdue invoice exposes a missing question. Could a Twin learn to ask it next time? A proposal for context that improves through evaluated, reversible changes.
- [I Built a Custom Blackwell Kernel for Qwen3.8-27B](https://mlx.systems/blog/qwen-custom-kernel-blackwell): I wanted more than 200 tokens per second from Qwen3.8-27B on one RTX PRO 6000. Getting there meant going below the serving framework.
- [China Is Winning the Post-Transformer Race](https://mlx.systems/blog/china-is-winning-the-post-transformer-race): I am bullish on Moonshot, Alibaba and Z.ai because they are releasing architectural progress that the rest of us can actually build on.
- [The Token Is a Narrow Feedback Channel](https://mlx.systems/blog/the-token-is-a-narrow-feedback-channel): A close reading of the full-bandwidth transformer: what latent feedback changes, what the 1B-parameter experiments establish, and why hidden computation cannot replace governed evidence.
- [RFQ1: Building a Relation-First Query Language for Project Venus](https://mlx.systems/blog/rfq1-relation-first-query-language): How RFQ1 represents relational queries, what its ten clauses mean, and how grammar-guided token masking can restrict a model to the language by construction.
- [Adapting Qwen3-VL Representations for Open-Set Continual Visual Learning](https://mlx.systems/blog/teaching-qwen-to-remember): A four-stage investigation into whether Qwen's internal visual representations could support few-shot, open-set continual learning.
- [You Are Not the Front Door](https://mlx.systems/blog/you-are-not-the-front-door): Why the user's chosen agent may become the primary entry point to software, and why every company will need a governed Twin between those agents and its systems.
- [Your AI Doesn't Need More Context. It Needs Approved Meaning.](https://mlx.systems/blog/approved-meaning-not-more-context): When sales, finance and delivery use the same words differently, retrieval alone cannot settle the report. Make the definition explicit, versioned and reusable.
- [Five Gates Between a Guess and a Governed Answer](https://mlx.systems/blog/five-gates-to-a-governed-answer): Follow a recurring-revenue definition through five distinct controls, and see why evidence from one cannot stand in for another.
- [One Twin, Many Interfaces](https://mlx.systems/blog/one-twin-many-interfaces): A practical map of how enabled interfaces reach the Twin, what stays consistent across them, and why a connection never grants itself wider access.
- [Every AI Answer Should Come With a Receipt](https://mlx.systems/blog/every-ai-answer-needs-a-receipt): A receipt is not a prose citation the model wrote about itself. It is a structured, machine-checkable output contract: product identity, execution identity, result identity, and an honest boundary marking exactly where proof ends.
- [A Connector Is Not Context](https://mlx.systems/blog/connector-is-not-context): Connecting Xero, Slack or SharePoint gives an AI system data access, not an agreed definition of your business. Here is how MLX separates connectors, plugins and governed Twin products.
- [The Agent Is the Easy Part](https://mlx.systems/blog/agent-is-the-easy-part): As the agent layer becomes easier to buy and replace, value moves toward the company context no vendor can supply: reviewed definitions, bounded access, and evidence that survives the model.
- [Your Agent Remembers You. Who Remembers the Company?](https://mlx.systems/blog/who-remembers-the-company): Why persistent agent memory is useful but insufficient for organisational truth, and why a company needs governed definitions that are committed rather than merely recalled.
- [A2A Solves How Agents Talk. It Does Not Solve Who Is Allowed.](https://mlx.systems/blog/talk-is-not-authority): Why agent interoperability needs more than authentication, and how signed subject identity, capability intersection, governed products, and receipts prevent a helpful agent from becoming a confused deputy.
- [Context Is a Security Boundary, Not a Token Budget](https://mlx.systems/blog/context-is-a-security-boundary): Why context assembly is an authorisation decision, how untrusted evidence becomes an attack surface, and why deterministic capability and data boundaries must sit outside the model.
- [A Skill Is a Hypothesis Until It Passes an Evaluation](https://mlx.systems/blog/a-skill-is-a-hypothesis): Why well-written agent skills can help, do nothing, or make performance worse, and how product-bound evaluations turn procedural guidance into evidence.
- [CubeSandbox Lands: A Useful Signal for Customer-Controlled Agent Execution](https://mlx.systems/blog/cubesandbox): What CubeSandbox could change for isolated agent execution, how to validate its vendor-reported claims, and why sandboxing remains separate from governed Twin data access.
- [Agentic Finance Workflows on SunSystems: A Forward-Deployed Pattern](https://mlx.systems/blog/agentic-workflows-on-sunsystems): How MLX uses read-only SunSystems evidence for governed turnover, balance, journal, debtor, creditor, and management-reporting products without claiming ledger mutation.
- [Agentic Finance Workflows on Xero: A CFO and Project-Lead Pattern](https://mlx.systems/blog/agentic-workflows-on-xero): How MLX syncs Xero evidence into the warehouse, publishes approved finance products through the Twin, and keeps Xero creation, approval, reconciliation, and sending explicitly out of scope today.
- [Kimi K2.6 Lands: Why Open-Weights Models Change the Finance AI Calculus](https://mlx.systems/blog/kimi-k2-6): What Kimi K2.6 means for finance-model evaluation, and why governed Twin contracts make provider comparisons more useful without implying a current MLX integration.
- [Letting an LLM Write SQL Against Your Warehouse Safely](https://mlx.systems/blog/warehouse-sql-safety): The two MLX warehouse paths: compiled read-only queries over an activated Twin product, and guarded model-authored SQL for explicit exploration.
- [What 'Customer-Controlled AI' Actually Means for Finance](https://mlx.systems/blog/customer-controlled-ai): Four practical questions for finance teams: where evidence lives, which model receives it, what an agent may do and who owns the definition.
- [Skills, Not Prompts: How Forward-Deployed Engineers Codify Finance Processes](https://mlx.systems/blog/skills-not-prompts): How MLX separates approved business meaning, operational skills, deterministic guardrails, and evaluations so finance workflows do not depend on a clever prompt.
- [On-Prem AI for Finance: A Practical Path for Regulated Teams](https://mlx.systems/blog/on-prem-ai-for-finance): How to assess a customer-operated finance AI deployment, agree responsibilities and prove one useful read-only workflow before expanding.
- [LLM-Agnostic by Design: Why Finance AI Shouldn't Be Locked to One Vendor](https://mlx.systems/blog/llm-agnostic-by-design): How MLX separates governed Twin products from model routing, what multi-provider support does and does not mean today, and why switching models still requires evidence.
- [Why Your Operational Reporting Is Lying to You](https://mlx.systems/blog/operational-reporting-disconnect): Why siloed ledger, pipeline, and operations data breaks reporting, and how a governed Twin turns fragmented evidence into approved, source-backed business products.

## Changelog

Dated product changes, newest first. Full entries at https://mlx.systems/changelog.

- 2026-11-06, [A graph of your business, workflows you can try and repair, and more of the Twin in Claude and ChatGPT](https://mlx.systems/changelog#2026-11-06-a-graph-of-your-business-and-workflows-you-can-try-and-repair): A new harness answers from a graph of your business held in memory, so totals and comparisons come back exact and fast. You can try a workflow on a real item without writing anything, and the Twin can repair an item that failed. Claude, ChatGPT and the CLI now reach much more of your Twin.
- 2026-11-01, [Discussions with files, answers that link to their sources, and records that save as you go](https://mlx.systems/changelog#2026-11-01-discussions-with-files-and-answers-that-link-to-their-sources): Channel discussions take files, images and quoted replies, an answer now links to the exact email, Slack message or Xero record it read, and records save each field as you change it. Claude, ChatGPT and the CLI sign in on your Twin's own page and can ask it to act.
- 2026-10-30, [Twin Desk 1.0.11: emails shared with you, a card for every email, and @Twin](https://mlx.systems/changelog#2026-10-30-twin-desk-1-0-11-emails-shared-with-you-and-a-card-for-every-email): On the Mac, emails your team shares with you now sit in Mail, each email in a thread has its own card, and "@Twin" under an email asks the Twin about it.
- 2026-10-28, [Twin Desk 1.0.10: ask with files, decide proposals, calmer Mail](https://mlx.systems/changelog#2026-10-28-twin-desk-asks-with-files-and-decides-proposals): Drop, paste or share files with a question, decide every kind of Twin proposal without leaving Desk, choose the Twin's model, and read Mail more calmly, on the Mac and the iPhone.
- 2026-10-25, [Azure SQL, BigQuery and Asana, and the PDF behind an answer](https://mlx.systems/changelog#2026-10-25-azure-sql-bigquery-and-asana-and-the-pdf-behind-an-answer): Connect Azure SQL, BigQuery and Asana, ask about Xero budgets, open the PDF behind an answer, and follow an answer's Trace beside it. Twin Desk began testing on the Mac and the iPhone.
- 2026-10-18, [Trace any figure to its records, and discuss what you share](https://mlx.systems/changelog#2026-10-18-trace-any-figure-and-discuss-what-you-share): Every figure in an answer, widget or pivot opens the records behind it. Share a record, channel or session with colleagues to discuss it on its own page, and work a shared mailbox together.
- 2026-10-11, [Twin's home, canvases, and reviews that say what Accept does](https://mlx.systems/changelog#2026-10-11-twins-home-canvases-and-reviews-that-say-what-accept-does): MLX now opens on Twin's home, the Twin makes canvases you can comment on, present and share, and every review says what accepting it does. Shared mailboxes, pinned links and customers you can mention arrived the same week.
- 2026-10-04, [A workspace that explains itself](https://mlx.systems/changelog#2026-09-29-a-workspace-that-explains-itself): New workspaces get one setup checklist, channels open on an overview, and anything named in the workspace shows what it is, so a new team reaches its first useful answer without a walkthrough.
- 2026-10-04, [A page for every customer, supplier and person](https://mlx.systems/changelog#2026-10-04-a-page-for-every-customer-supplier-and-person): Each customer, supplier and person in your connected systems now has a page that gathers what your systems hold about them, so you stop piecing one together from several systems.
- 2026-10-04, [Workflows the Twin writes for you, with reviews on the record](https://mlx.systems/changelog#2026-10-04-let-the-twin-write-workflows): Describe a workflow that should run on each new document or changed record, and the Twin drafts it, so recurring checks become something you describe rather than build. Its reviews now wait on the record they're about.
- 2026-10-04, [Sessions you can steer, in a calmer workspace](https://mlx.systems/changelog#2026-10-04-sessions-you-can-steer-in-a-calmer-workspace): Steer the Twin while it works: queue a follow-up, roll back to an earlier turn, or answer its question. The workspace around it is calmer and more consistent.
- 2026-10-04, [In testing: Twin Build learns what your source codes mean](https://mlx.systems/changelog#2026-10-04-twin-build-in-testing-reads-what-source-codes-mean): Twin Build can now build data products for a precise slice of your data, such as sales invoices only, and the Twin answers questions about a specific customer or supplier in fewer steps.
- 2026-10-04, [In progress: guests in your channels](https://mlx.systems/changelog#2026-10-04-work-in-progress-guests-in-your-channels): We're building guest access, so you can bring people from outside your organisation, such as clients, into a channel. It isn't available yet.
- 2026-09-30, [Follow an answer's path step by step](https://mlx.systems/changelog#2026-09-30-follow-an-answers-path-step-by-step): When a number matters, you can now see exactly how the Twin reached it: an answer's trace opens each step beside the canvas, and every canvas scrolls and zooms like a design tool.
- 2026-09-30, [Schedule workflows to run on their own](https://mlx.systems/changelog#2026-09-30-schedule-workflows-to-run-on-their-own): Workflows can now run on a schedule as their owner, post each result to a channel and list every run, so the weekly report writes itself and lands where your team already reads it.
- 2026-09-29, [In testing: a harness that offers the right data more often](https://mlx.systems/changelog#2026-09-29-in-testing-a-harness-that-offers-the-right-data): Harness v3 decides which Data Sets the Twin may use for each question. It's in testing and off for customers.
- 2026-09-27, [Build and watch workflows on a canvas](https://mlx.systems/changelog#2026-09-27-build-and-watch-workflows-on-a-canvas): Repeatable work you can see and trust: workflows gained a visual canvas for building steps and following each run, and dashboards gained date filters.
- 2026-09-20, [See the evidence beside the answer](https://mlx.systems/changelog#2026-09-20-see-the-evidence-beside-the-answer): Checking an answer became one click beside the conversation: Sessions gained one pane for the sources, documents and evidence behind it. Calendars became editable, and workflows arrived in early preview.
- 2026-09-13, [Answers with verifiable evidence](https://mlx.systems/changelog#2026-09-10-answers-with-verifiable-evidence): Answers with tables and figures now tie every number to the exact query result it came from, so you can check them before relying on them.
- 2026-09-13, [A Twin hub, richer links and channel files](https://mlx.systems/changelog#2026-09-13-a-twin-hub-richer-links-and-channel-files): Settings moved into one Twin hub, pasted links show what they point to, and a channel's files sit beside its work, so setup takes fewer steps and nothing hides behind a bare URL.
- 2026-09-06, [One place to inspect what your Twin knows](https://mlx.systems/changelog#2026-09-06-one-place-to-inspect-what-your-twin-knows): Knowledge brings your files, the Twin's own work, connected sources, Data Sets and Twin Build into one place, so you can see what your Twin knows, and where each piece came from, before you rely on it.
- 2026-08-30, [Twin in workplace conversations](https://mlx.systems/changelog#2026-08-30-bringing-twin-into-workplace-conversations): Use the Twin from Slack, Microsoft Teams and Microsoft 365 Copilot, with the same permissions everywhere, and see everything that needs you in one Inbox.
- 2026-08-23, [Dashboards you can work with](https://mlx.systems/changelog#2026-08-23-dashboards-you-can-work-with): Dashboards became interactive: filter from a chart, rearrange widgets and go from a headline number to the records behind it without exporting to a spreadsheet.
- 2026-08-21, [Answers stream with their working](https://mlx.systems/changelog#2026-08-21-answers-stream-their-working): Watch the Twin work: the answer writes in front of you, and the evidence beneath it fills in as each source is read.
- 2026-08-16, [Memory you can inspect and control](https://mlx.systems/changelog#2026-08-16-building-memory-you-can-inspect-and-control): We began building Memory, context the Twin keeps between Sessions, with every version, source and use visible, so it can help without becoming a black box.
- 2026-08-09, [Keep the conversation beside the data](https://mlx.systems/changelog#2026-08-09-data-sets-become-shared-working-spaces): Data Sets became shared working spaces: one set of records, many ways to view it, and the discussion right beside it, so decisions stay with the record instead of scattering across email and chat.
- 2026-08-02, [See how current your data is](https://mlx.systems/changelog#2026-08-02-see-how-current-your-data-is): Every connected Data Set now shows whether its data is current, refreshing or stale, so you know whether a number is safe to act on before you act on it.
- 2026-07-26, [Bring Twin into approved tools](https://mlx.systems/changelog#2026-07-26-bring-twin-into-approved-tools): Use the same Twin from other assistants and developer tools, with one set of permissions and one history, wherever your team prefers to work.
- 2026-07-19, [Connect approved Postgres data](https://mlx.systems/changelog#2026-07-19-connect-approved-postgres-data): Ask questions of an approved Postgres database directly, without a data project to copy it somewhere else first.
- 2026-07-12, [Saved Views for everyday work](https://mlx.systems/changelog#2026-07-12-saved-views-for-everyday-work): Save how you like to see a channel's records, as a table, board, calendar or timeline, so everyone returns to the same view of the work instead of rebuilding filters.
- 2026-07-05, [Private Channels and clearer evidence](https://mlx.systems/changelog#2026-07-05-private-channels-and-clearer-answer-evidence): Keep sensitive work to named members, inside the same workspace, and see which data each answer queried.
- 2026-06-28, [Documents in Sessions, first steps on iPhone](https://mlx.systems/changelog#2026-06-28-documents-in-sessions-and-an-iphone-foundation): Attach a document to a Session and the Twin works from it, instead of you copying its contents across, and an early iPhone app joined the same workspace.
- 2026-06-21, [Twin conversations in Teams](https://mlx.systems/changelog#2026-06-21-a-teams-path-into-twin): Start work with the Twin from a Microsoft Teams conversation and get the result back in the same thread, without switching tools.
- 2026-06-14, [One workspace across web and Mac](https://mlx.systems/changelog#2026-06-14-one-workspace-across-web-and-mac): The web and Mac apps began sharing the same Sessions, Channels and workspace, so work continues wherever you open it.
- 2026-06-07, [Channels, Inbox and Sessions come together](https://mlx.systems/changelog#2026-06-07-channels-inbox-and-sessions-come-together): The workspace took its current shape: Channels for shared records, the Inbox for what comes in, and Sessions for your work with the Twin, in one place instead of three tools.
- 2026-05-31, [Proposed changes with an audit trail](https://mlx.systems/changelog#2026-05-31-proposed-changes-with-an-audit-trail): An early review flow let people inspect a suggested change, with its evidence, before anything was applied, so nothing changed without a person seeing why.
- 2026-05-24, [Clearer setup for business tools](https://mlx.systems/changelog#2026-05-24-clearer-setup-for-business-tools): Connecting a business tool became a guided flow that tells you what's ready and what still needs attention, so setup takes minutes rather than a support ticket.
- 2026-05-17, [More dependable connected data](https://mlx.systems/changelog#2026-05-17-more-dependable-connected-data): Imports from connected systems became resumable, so a dropped connection or a provider's rate limit no longer means starting again, and your figures stay current.
- 2026-05-10, [Work you can stop and continue](https://mlx.systems/changelog#2026-05-10-work-you-can-stop-and-continue): Conversations with the Twin became lasting: answers are saved, you can stop work mid-run, and a follow-up picks up where you left off instead of starting over.
- 2026-05-03, [Every query stays inside your organisation's data](https://mlx.systems/changelog#2026-05-03-stronger-boundaries-for-connected-data): Every query the Twin runs is now checked against the data its task is allowed to use before it runs, so its work stays within what your organisation allows.
- 2026-04-26, [Reusable Skills, clearer progress](https://mlx.systems/changelog#2026-04-26-reusable-skills-and-clearer-progress): Teach the Twin a way of working once, as a Skill, and every run can follow it, so good practice spreads without retraining anyone.
- 2026-04-19, [Start work with a prompt](https://mlx.systems/changelog#2026-04-19-a-workspace-that-starts-with-a-prompt): The first version of the workspace: describe what you need, and the task, its progress and its result stay together in one conversation you can come back to.

## Links

- Website: https://mlx.systems
- Accountants: https://mlx.systems/for/accountants
- Consultancies: https://mlx.systems/for/consultancies
- Creative & media: https://mlx.systems/solutions/creative-media
- Private equity & venture capital: https://mlx.systems/solutions/pe-vc
- Technology companies: https://mlx.systems/solutions/technology
- Fintech: https://mlx.systems/solutions/fintech
- Food & consumer goods: https://mlx.systems/solutions/fmcg
- Hotels: https://mlx.systems/solutions/hotels
- Fractional finance & operations: https://mlx.systems/solutions/fractional
- Documentation: https://docs.mlx.systems/docs
- Changelog: https://mlx.systems/changelog
- Blog: https://mlx.systems/blog
- Blog feed (RSS): https://mlx.systems/blog/rss.xml
- Full blog corpus (markdown): https://mlx.systems/llms-full.txt
- Per-post markdown: https://mlx.systems/blog/{slug}/markdown
- Project Venus research: https://mlx.systems/research
- Case studies: https://mlx.systems/case-studies
- About: https://mlx.systems/about
- Support: https://docs.mlx.systems/docs/support
- Pricing: https://mlx.systems/pricing
- FAQ: https://mlx.systems/faq
- Sub-processors: https://mlx.systems/subprocessors
- Connect Stripe with Twin Sync: https://mlx.systems/connect/stripe
- Install Twin for Slack: https://mlx.systems/connect/slack
- Connect Xero: https://mlx.systems/connect/xero
- Legal & Security: https://docs.mlx.systems/docs/policies
- Privacy: https://docs.mlx.systems/docs/policies/privacy
- Terms: https://docs.mlx.systems/docs/policies/terms

## Contact

- General: team@mercurylabs.io
- Sales: team@mercurylabs.io
- Support: team@mercurylabs.io