Hi HN — we’re Aryah and Ajay, the founders of o11 (o11.ai).
We build AI agents that run inside Microsoft 365 and Google Workspace apps so you can create and modify files directly in PowerPoint, Excel, Word, Slides, Docs, and Sheets.
The problem we kept running into was simple: most real work still happens in “legacy” productivity software, but most AI tools live outside of it. You paste data into a chat interface, get text back, and then manually move it into your slides or spreadsheets. Even the built-in AI features in tools like Microsoft Copilot or Gemini tend to generate suggestions rather than actually carrying out multi-step work inside the file.
Last year, before leaving University of North Carolina at Chapel Hill, we saw finance teams spending hours building slide decks and updating Excel models by hand during internships. We tried existing AI tools, but they mostly produced text snippets. The friction wasn’t in writing a paragraph — it was in assembling and formatting the actual file.
So we started with a PowerPoint add-in. Instead of generating text in a sidebar, it can construct and modify slides directly using the native Office APIs: create layouts, insert tables, format charts, update numbers, restructure content. Users then asked for Excel support, then Word. After that we built the same capability set for Google Slides, Docs, and Sheets.
Under the hood, we treat each host app as an execution environment. The model doesn’t just return text; it generates structured actions that map to the app’s object model (e.g., slide trees, cell ranges, document blocks). We maintain a state representation of the file so the agent can reason about what already exists before making changes. That helps avoid the “AI overwrote my document” problem and makes edits incremental rather than destructive.
There’s a free tier to try o11. For Launch HN, we’ve removed demo gates — you can install and use it directly.
Where we’re heading is broader than Office and Workspace. Our hypothesis is that most enterprise software will remain in place for a long time, and automation will need to operate inside those interfaces rather than replacing them. We’re working on a generalized runtime that lets the same agent framework execute actions in other desktop and web apps.
We’re still early and would genuinely appreciate feedback — especially from people who’ve built on Office.js, Google Apps Script, or other plugin ecosystems. We’re also happy to answer technical questions about how we handle file state, permissions, or reliability inside these host environments.
aoztanir•2h ago
We build AI agents that run inside Microsoft 365 and Google Workspace apps so you can create and modify files directly in PowerPoint, Excel, Word, Slides, Docs, and Sheets.
The problem we kept running into was simple: most real work still happens in “legacy” productivity software, but most AI tools live outside of it. You paste data into a chat interface, get text back, and then manually move it into your slides or spreadsheets. Even the built-in AI features in tools like Microsoft Copilot or Gemini tend to generate suggestions rather than actually carrying out multi-step work inside the file.
Last year, before leaving University of North Carolina at Chapel Hill, we saw finance teams spending hours building slide decks and updating Excel models by hand during internships. We tried existing AI tools, but they mostly produced text snippets. The friction wasn’t in writing a paragraph — it was in assembling and formatting the actual file.
So we started with a PowerPoint add-in. Instead of generating text in a sidebar, it can construct and modify slides directly using the native Office APIs: create layouts, insert tables, format charts, update numbers, restructure content. Users then asked for Excel support, then Word. After that we built the same capability set for Google Slides, Docs, and Sheets.
Under the hood, we treat each host app as an execution environment. The model doesn’t just return text; it generates structured actions that map to the app’s object model (e.g., slide trees, cell ranges, document blocks). We maintain a state representation of the file so the agent can reason about what already exists before making changes. That helps avoid the “AI overwrote my document” problem and makes edits incremental rather than destructive.
There’s a free tier to try o11. For Launch HN, we’ve removed demo gates — you can install and use it directly.
Where we’re heading is broader than Office and Workspace. Our hypothesis is that most enterprise software will remain in place for a long time, and automation will need to operate inside those interfaces rather than replacing them. We’re working on a generalized runtime that lets the same agent framework execute actions in other desktop and web apps.
We’re still early and would genuinely appreciate feedback — especially from people who’ve built on Office.js, Google Apps Script, or other plugin ecosystems. We’re also happy to answer technical questions about how we handle file state, permissions, or reliability inside these host environments.
Thanks for reading.