Averda AI

API endpoint API key (optional) Context window (tokens)

The Local Ollama endpoint /ollama is proxied by the dev server to http://localhost:11434, avoiding CORS issues. Endpoints may include /v1 or not — both work.
Recommended model: gemma4:e4b (tools, thinking, 128k max context) or granite3.3 (tools) — small models (≤4B) may ignore formatting rules or produce malformed actions. Whichever model you pick, Ollama needs a bigger context: in the Ollama app set Settings → Context length → 16k+ and restart Ollama (or OLLAMA_CONTEXT_LENGTH=16384 for ollama serve; see Help & About), or replies will be cut short.

Setup

Hosted (managed): sign in with your Microsoft work account when prompted. Models are configured by your Averda admin — pick one in the composer dropdown; a green dot means connected.
Local development: run ollama serve and npm run dev, then open ⚙ → Connection, pick Local Ollama or Averda AI Gateway, and press Save & test.

How to use

Select cells in your sheet — the selection is attached automatically (shown as 📎 next to the input). Use ↗ to pin a specific range.
Type your request and press Enter — e.g. "summarize this table", "add a total row", "make this a proper table".
When the assistant proposes workbook changes, review the action cards and click Apply (writes into cells that contain data need a second click to confirm the overwrite).
Use ■ to stop a long answer, ＋ to start a new chat, and the model dropdown in the composer to switch models.

What it can do

Read and write ranges, fill ranges, insert formulas, format cells, create formatted tables, set number formats, sort, auto-fit columns, create sheets and charts — always with your confirmation.

Ollama context window

Ollama runs models with a 4,096-token context by default — too small for this assistant, so replies get cut short. Give models a 16k window once:

Ollama app: menu-bar icon → Settings → Context length → 16k or more, then Quit Ollama and open it again (launchctl setenv does not work — the app keeps its own setting).
Terminal/brew ollama serve: set OLLAMA_CONTEXT_LENGTH=16384 in that environment and restart the server.
Or run npm run ollama:context in the project folder — it checks and applies this for you.
Verify after the first chat: ollama ps → CONTEXT 16384.
Alternative for one model: ollama create averda-gemma -f Modelfile with FROM gemma4:e4b and PARAMETER num_ctx 16384.

Troubleshooting

Certificate error: run npx office-addin-dev-certs install, then fully quit and reopen Excel.
Red status dot: check Ollama is running, or verify the gateway URL and key under ⚙ → Connection.
Stale pane: pane menu (ⓘ) → Reload or Clear Web Cache, then reopen the pane.

About

Averda AI for Excel
An Office.js add-in that connects Excel to private, OpenAI-compatible LLM endpoints (local Ollama by default, or Averda AI gateway).

Averda AI for Excel is proprietary software. © 2026 Averda. All rights reserved. This tool and its source code are the intellectual property of Averda and are covered under Averda's IP rights. Unauthorized copying, distribution, or use outside Averda is prohibited.