AI Assistant: bring your own API key

TaskRely's AI features run on an AI provider you choose, using your organization's own API key. There's no AI surcharge on your TaskRely plan and no credits to buy - you pay your provider directly, at their prices, and you can switch providers whenever you like.

How it works

An organization admin connects one AI provider for the whole organization, the same way Google Drive is connected. From then on, every member the admin allows sees small ✨ buttons next to the tools they already use - the description editor, subtasks, the board, reports and diagrams.

AI in TaskRely only ever suggests. Every result is shown to you first, and nothing is saved to your workspace until you click Apply, Add, Create or Save - the same buttons you'd use if you'd done it by hand. If your organization never connects a provider, TaskRely works exactly as it always has.

Connecting a provider

Go to Integrations in the sidebar and click Connect AI provider on the AI Assistant card (admins only). Then:

• Pick a provider - Anthropic, OpenAI, Google Gemini, Groq, or any OpenAI-compatible service (OpenRouter, Together, a self-hosted vLLM or Ollama server). For a compatible service, enter its base URL; it must be reachable over https from the internet - a server on your own laptop (localhost) won't work.
• Pick a model - choose a suggestion or type any model id your provider offers.
• Paste your API key and confirm you're happy for task content to be sent to that provider.
• Click Test & connect. TaskRely makes one tiny test call first - if the key, model or URL is wrong you'll see why, and nothing is saved.

Want to try it for free first? Groq offers a free API tier - create a key at groq.com, choose Groq as the provider, and connect.

The key is encrypted when it's stored and is never shown again - the card only displays its last four characters. To switch model or provider later, click Change; you can leave the key blank to keep the current one, as long as the provider stays the same.

What you can do with it

Ask AI - the ✨ button at the top of the sidebar - is the assistant that looks things up and does the clicking for you. Ask it questions or for reports ("Create a report about Jenny's project distribution", "Which tasks are overdue, and who owns them?", "Who is on leave next week?") and it answers from your live data - with copy and download - or describe what you want ("Create a project RelyTask with 4 statuses and departments Content, SEO and AI", or "In Website, add a task Fix login to Backlog, assign Rahul, due Friday") and it plans the steps: creating projects, adding tasks with assignees, deadlines and priorities, updating tasks, and adding subtasks. You review the plan first - anything it couldn't match, like a person who isn't on the project, is flagged - and only when you click Do it does it carry the steps out, one by one, then shows what was done. It never deletes anything, and moving a task to a Done column stays on the board so completion checks still run.

Right where you work, you'll also find:

• Write & improve descriptions - the ✨ AI menu beside a task description's Edit button can write a first draft, improve, shorten, change the tone or translate it. The draft opens in the normal editor so you can adjust it before you save.
• Suggest subtasks - ✨ Suggest next to Add subtask proposes a breakdown; tick the ones you want and add them in one go.
• Catch me up - above a task's comments, get a short recap of the discussion: decisions, open questions and next steps.
• Suggest priority, size & labels - picks only from your project's own priorities, sizes and labels, with a one-line reason. Label suggestions tick the boxes in the labels picker; close it to apply.
• Plain-English quick add - the ✨ button beside a board column's Add item turns "Rahul to fix the login bug by Friday, high priority" into a task with the assignee, due date and priority filled in. You see a preview before it's created.
• Project status summary - on a project's Reports page, write a standup-style update of what's done, in progress, blocked and worth focusing on. The counts are calculated by TaskRely, so they're always exact.
• Flowcharts & mind maps - the ✨ AI button in the diagram editor builds a flowchart or mind map from a description, or from the project's tasks. It's added next to what's already on the canvas - nothing is replaced - and one Undo removes it.

Who can use it

AI is available on the Growth and Business plans. Admins control it from Settings on the AI Assistant card:

• Features - switch each AI feature on or off for the whole organization.
• Access - choose which roles can use AI, then override individual people with Always allow or Block. Admins always have access.
• Monthly token cap - optional; once the organization uses that many tokens in a calendar month, AI pauses until the next month.

Each person can make up to 30 AI requests every 10 minutes. The card shows this month's requests and tokens, so you can see usage building up before it becomes a surprise.

Privacy & costs

When someone uses an AI feature, only what that request needs is sent to your provider - for example a task's title, description and comments, or a project's task list. AI can only see what that person could already see in TaskRely; a project they don't have access to is never included. TaskRely keeps a count of requests and tokens, not the prompts or answers themselves.

Usage is billed by your provider to your key. Smaller, faster models are usually plenty for these features and cost a fraction of the largest ones.

Troubleshooting

• The ✨ button is greyed out - hover it to see why: your plan doesn't include AI, no provider is connected, that feature is switched off, or your admin hasn't enabled AI for you.
• "The AI provider rejected the API key" - the key was revoked or expired; an admin can paste a new one with Change.
• "Rate limiting or out of quota" - your provider's own limit was reached (common on free tiers). Wait a moment, or raise the limit with your provider.
• "An answer in an unexpected format" - try again, or switch to a more capable model; very small models sometimes struggle with structured results.