FrontiersOS
The in-house AI platform Frontiers runs on.
What is FrontiersOS?
FrontiersOS is the in-house AI platform at Frontiers, one of the world’s largest open-access research publishers. It puts AI agents inside the systems the company already runs — the Publishing Platform, the data warehouse, Salesforce, Microsoft 365 — and replaces none of them.
The bet behind it: output should stop scaling with headcount. Work already happens everywhere — an inbox, a review forum, a CRM tab, a warehouse query — so the agents live everywhere too, and take the repetitive half. What is left is the half only people can do: strategy, new markets, the relationships that move science.
And the people closest to the work are the ones who build for it. Everything below is self-service: an editor, an analyst or a marketer assembles their own agents, skills, automations and apps, with no engineering queue to wait in and humans in and on the loop throughout.
What’s in it
Three layers: the blocks you assemble into a solution, the data and context agents read from, and the places you reach them.
Building blocks
What you assemble into a solution.
- Agents
- A named chat config — instructions, model and tools — that holds role context so you stop re-explaining it.
- Skills
- Versioned knowledge bundles synced from a git repository, auto-loaded when a conversation needs them.
- Apps
- A purpose-built UI over an agent, its skills and its tools, for a job the whole team does.
- Automations
- A prompt and an agent that run when a trigger fires — a schedule, a webhook, an inbound email.
- Workflows
- Deterministic pipelines, callable as tools where no model judgement belongs.
- Commands
- Text shortcuts that expand anywhere you type, in the browser or a native app.
Data & context
What agents read from and remember.
- Datasets
- A read-only catalogue over the data warehouse — the funnel, revenue, reviewers.
- Databases
- A personal SQL database wired into every workspace, for analyses that outlive one chat.
- Embeddings
- Vector indexes for semantic search: find similar reviewers, articles, topics.
- Memory
- Durable per-user facts, injected up front across every conversation.
- Models
- The catalogue of licensed models, routable per chat or per @mention.
- Workspaces
- Ephemeral cloud VMs per chat, for code, files, browser work and generated reports.
Access surfaces
Where you reach it.
- Extension
- The browser side-panel and in-page commands — the primary daily surface, on any site.
- Desktop
- A native app that expands command shortcuts in Outlook, Teams, Word and editors.
- Mobile
- An iOS app with chat, photos, per-chat tools and push notifications.
- Teams bot
- The assistant inside Microsoft Teams, where a lot of the work already happens.
- Widget
- An embeddable chat, backed by a pinned agent, for an external web page.
- MCP server
- The platform exposed over the Model Context Protocol to outside clients.
The shipping record
Every one of these is written up in full on the releases page.
- Releases shipped
- 56
- Shipping since
- March 2026
- Latest version
- 2.12.5
- Built by
- Innovation team
How it’s built
- Typed end-to-end web application
- Server-side rendering
- Multi-provider LLM routing
- Agent tool-calling framework
- Retrieval-augmented generation
- Vector search
- LLM evaluation and tracing
- Relational database with migrations
- Message queues
- Durable workflow orchestration
- Ephemeral cloud sandboxes
- Container orchestration
- Enterprise single sign-on
- Cloud data warehouse
- Structured logging and error monitoring
- Browser extension and native clients
- Embeddable widget and public API
Built by the Innovation team
FrontiersOS is designed, engineered and run end to end by the Innovation team at Frontiers — architecture, backend, frontend, infrastructure, model orchestration, evaluations and every access surface above.
Main contacts
Want to build things like this with us? See open positions at Frontiers