Roadmap
Agent OS, coming in 6.0
Seven parts, one story: everything your AI agents need to operate as one governed team inside your company.
Specialized agents that share context, skills and tools — each with a page of its own, running in our cloud or on your own server.
Everyone works on the same company data, and every person and agent sees exactly what they were given.
A skill one agent learns becomes the whole team's, and a curated, security-scanned library adds more.
Decide which accounts and actions each agent can use, and see every connection in one place.
The right model or data source for every task, paid from one balance your whole team shares.
Thousands of records processed in parallel, in our cloud or on your agents' own servers — and every run in one place.
The work delivered, and the time and money it saved — estimated with assumptions you set.
In development
The whole app redesigned around your agents and your data, arriving with 6.0.
An agent changing a whole selection of records in one pass instead of one call at a time.
Leaner responses and lighter loading where a system holds hundreds of thousands of records.
The four-language family completes, from the same generated pipeline.
Next up
Hand it the work by voice or by text, and it splits the job between your agents.
Talk to any agent right in the app, with dictation and read-aloud.
A report or a board an agent builds from your data, hosted here and living inside your permissions.
Exploring
Say what must stay true — never oversell, one clean record per customer — and get proof it stayed true.
Small writes flow; big or irreversible ones wait with a dry-run diff and their blast radius.
A proven schema with its automations and views, fitted by an agent to the data you already have.
Every change an agent made to your data on record, and any of it one step from undone.
Answer a question about your data once, and it becomes a rule you are never asked about again.
Shipped recently
| Feature | What it does |
|---|---|
| Much faster tables | Sorting, filtering and resizing redraw only what changed, and large modules open with less work for your browser |
| Code automations | Your own Python as an automation step, in a sandbox — written by you, or by an agent that tests it against your data before it ships |
| More capable MCP tools | Connected agents now build structure too — modules, spaces, field groups, fields — put files on records, and ask for a secret through a form instead of the conversation |
| Rollup, Formula and button fields | A value aggregated across a record's linked records, one computed from the fields beside it, and an action button on the record |
| Vector database | Spaces embedded for meaning, so retrieval works by intent — on selected systems |
| Build a module with AI | Describe what you want to track; AI generates the module and its typed fields |
| TypeScript and Python SDKs | Typed clients generated from the REST API spec |
The full record lives in the release notes for what shipped in the product, the API changelog for every public API change, the MCP changelog for the server agents connect to, and the SDK changelog for client releases.
Direction, not delivery dates — priorities follow what customers ask for. Waiting on something above? Tell us.
