Platform Release · Summer 2026
V27 is the release where the platform becomes something you direct rather than operate.
A crew of AI specialists you lead by conversation.
A home for every kind of data — to learn from, process, and query.
A no-code way to build the apps your teams need.
A layer of trust over every action.
Four capabilities, one platform.
V27 · by the numbers
Experience
One idea quietly rewrites delivery economics this cycle: one rendering engine assembles many different applications from a specification. No bespoke front-end per customer. Cards are the building blocks; cards compose into Pages; Pages bundle into Packs — a whole app layered over the base product.
Each card is screen- and data-aware but free of any single business's specifics — so a card built for a lending workbench works unchanged on a claims desk.
Cards wire to your data model through configuration, with a visual designer and live preview. Adding a brand-new card is a registry entry, with zero edits to the renderer.
This is not a preview — whole solutions run on this layer today:
A source viewer beside a two-period spread grid, per-line verify and flag, ratios, covenants and a decision panel.
Documents drop in, classify, split and extract with provenance, then self-fill a form dozens of fields deep.
Assembled entirely by configuration — metric cards, charts and structured-data lists, with live refresh.
A tailored, customer-specific application now takes days, by configuration — and it travels between environments as a single portable package.
Piper
Piper is one interface that navigates the platform, configures it, runs it, and reports back — in plain language. It does not work alone: behind Piper stands a crew of five specialists it directs.
You lead Piper; Piper leads the crew — routing each ask to the right specialist, with over 200 platform actions at its command.
Every specialist is scoped to its role. Ask it to read your data, and reading is all it does — changes and deletions run through the specialists and approvals built for exactly that.
The agent factory
Underneath the copilot, Botminds is an agent factory. Six agent types assemble from the same building blocks.
Pulls structured fields from any document.
Categorizes and routes — by section, page, or whole document.
Normalizes messy values into canonical forms.
A first-class reviewer that checks another agent's work before you trust it.
Grounds chat in your own documents.
Orchestrates a crew of member agents, shown live in the Crew view.
Tools and skills load into context on demand — an agent stays fast and lean no matter how large your library grows.
Mesh
Directing a crew is half the story; running it at enterprise volume is the other. The Mesh execution engine was re-architected this cycle: dispatch is decoupled from execution, so the ceiling is your real worker pool.
Dispatch decoupled from execution — 2.8 runs per second before, 11.2 after. Large batches clear dramatically faster on the same hardware.
Long, human-in-the-loop work now parks and resumes instead of holding a worker hostage.
Every document, stage and agent moving in real time — you watch work flow through the platform as it happens.
Collections
Data comes in three shapes — a library to learn from, files to process, and tables to query. You pick a Purpose when you create a collection, and the platform shapes itself around it.
A corpus your agents learn from. Automatically indexed, kept fresh, protected from accidental deletion, and attachable to any agent for retrieval grounded in exactly your material.
Documents flow through stages while agents act on them, on a lifecycle you control, with a built-in two-approval "four-eyes" workflow before anything is final.
A rate card, a rulebook, a catalog kept as a real table: typed columns you query, sort and edit — and still a full collection with workflow, agents, review and dashboards behind it.
Typed columns you query, sort and edit, a Reference field that links records into connected master data, all defined in the no-code Schema Designer — with agent-powered self-enrichment behind it.
Your data keeps its natural shape: a knowledge base behaves like a knowledge base, a pipeline like a pipeline, a table like a table — each with the right indexing, workflow and behavior by default.
Governance
In V27, control comes built in. Governance is a first-class pillar of Studio, sitting beside Agents, Data and Experience — backed by mechanism you can point to.
The copilot that acts on your platform also asks: destructive and sensitive operations route through explicit human approval.
Every action lands in the trail — who did what, on which object, under which policy.
Cost is visible where the work happens, so you ship under control and see every dollar.
The full list
The keynote above is the story — this is the changelog: 700+ enhancements across seven areas, item by item.
The shape of it
That's V27.