Compatibility

OpenDocBot is a single add-in that runs in Word, Excel and PowerPoint. One add-in, all three apps, no separate installs.

Is your Office version supported?

Platform	Support
Microsoft 365 (desktop)	✅ Fully supported
Microsoft 365 (web)	✅ Fully supported
Office 2016/2019/2021 (perpetual)	❌ Not supported
LibreOffice / Google Docs	❌ Not supported

Required API sets (for IT administrators)

OpenDocBot targets Word, Excel, and PowerPoint. Because each host exposes different API sets, host-specific sets (WordApi, ExcelApi, PowerPointApi) are not declared globally in the manifest. They're checked at runtime per host with isSetSupported. The only global requirement in the manifest is the shared runtime:

API set	Min version	Used for
SharedRuntime	1.1	Persistent taskpane runtime

At runtime the add-in requires, per host:

Host	API set	Min version
Word	WordApi	1.3
Excel	ExcelApi	1.1
PowerPoint	PowerPointApi	1.4

Advanced PowerPoint editing needs 1.8

Most PowerPoint features work with PowerPointApi 1.4, but advanced slide editing requires PowerPointApi 1.8. Current-channel Microsoft 365 has it. If an advanced editing tool fails, check your channel version.

AI Provider compatibility

OpenDocBot is modular and supports several AI providers. Under the hood there are three providers: OpenAI-compatible (used by OpenAI, DeepSeek, Ollama, OpenRouter and any Custom OpenAI-compatible endpoint), Anthropic (Claude), and Gemini (Google).

A preset is just a shortcut: a provider bound to a specific endpoint and configurations. Several presets share the same provider (for example, Ollama and OpenRouter both use the OpenAI-compatible provider). Because presets reuse providers, you can switch between them without changing anything else.

Not all features are available with every provider.

Feature matrix

Feature	OpenAI	DeepSeek	Anthropic	Gemini	Ollama	OpenRouter
Reasoning	✅	✅	✅	✅	✅*	✅*
Reasoning effort	✅	✅	✅	✅	❌**	⚠️**
Token streaming	✅	✅	✅	✅	✅	✅
Prompt caching	✅	✅	✅	✅	❌	✅
Cache TTL options	-	-	5m / 1h	rebuild size	-	-
Legacy /chat/completions toggle	✅	✅	n/a	n/a	✅	✅
Model list fetching	✅	✅	✅	✅	❌	✅

* Reasoning availability depends on the underlying model; an endpoint only surfaces reasoning if the model emits it (e.g. reasoning_content for DeepSeek-style models).

** Reasoning effort depends on the model and endpoint. OpenAI/Anthropic/Gemini send it natively; Ollama ignores it, and OpenRouter forwards it to the routed model, which may or may not support it.

Notes

• The Ollama preset needs no API key, so the model name is typed manually. If Ollama rejects requests, toggle Use old /chat/completions endpoint in Advanced (some versions don't support /responses).
• The Gemini provider supports context caching with a configurable rebuild threshold (default 2000 tokens); the cache is deleted on taskpane close to stop idle-storage billing.
• The OpenRouter preset lets you filter the model list to show only :free models, so you can use the add-in at no cost.

CORS and the proxy

The add-in runs in your browser (the Office taskpane is a webview), so it can only reach providers that allow browser-origin requests (CORS).

To connect to providers not supporting CORS (e.g. OpenCode), a request proxying functionality was added. This functionality is only available on self-hosted instances (see Self-hosting). OpenCode has no preset; configure it through the Custom preset with the proxy enabled (see OpenCode).

Provider	Hosted instance	Self-hosted
OpenAI	✅	✅
DeepSeek	✅	✅
Anthropic	✅	✅
Gemini	✅	✅
Ollama	✅ (any endpoint)	✅
OpenRouter	✅	✅
OpenCode	❌	✅ (via proxy)

Request proxying can be turned ON/OFF via Settings → Connection → Advanced → Proxy API requests through this server.