What does using a model with your own key mean in Task Manager?
Task Manager doesn't sell its own AI. Connect a model from one of four providers - OpenAI, Google's Gemini, Anthropic's Claude or xAI's Grok - using your own API key. Longer responses may increase the cost of using the model.
Get an API key from your chosen provider.
The key is encrypted and never shown in full after you save it. Changing other AI settings doesn't remove the key. You can enter a new key when changing the configuration.
What does connecting a model involve?
You can open AI settings using the cog icon in the top left, under the AI tab. Select a provider from the “AI provider” list and paste in your key. Then click the refresh button (↻) next to the model list. Task Manager fetches the latest list from the provider and shows models that can hold a conversation. This isn’t a fixed list of models built into the programme. The (recommended) label is currently shown for gpt-4o, gemini-2.0-flash, claude-opus-4-8 and grok-3.
After clicking “Save AI settings”, the application tests the connection with the chosen model and shows warnings if it detects a problem. The application smooths over differences in request formats between providers and models on its own, without any effort on your part.
Semantic search in the knowledge base (Pro plan), i.e. searching by meaning, only works with OpenAI and Gemini. For them it's enabled by default. You set the maximum response length within the range of 1024-16384 tokens; the default value is 4096. A higher limit allows for longer responses and increases the cost with the provider.
What do you gain?
You can change the model when the current one no longer suits your team’s work. If a cheaper model is enough for summaries, you don’t have to wait for the programme’s developers to add it to their offering. You can also try a new model made available on your account.
You don't have to go through every old task and chat to get back to decisions and arrangements - in the Pro plan, the AI knowledge base keeps their summaries, and fragments of the original content from notes and files.
In the Pro plan, your team can search for earlier arrangements by meaning, not just by keywords, if it uses OpenAI or Gemini.
When is it worth it? An example
A design studio on the Pro plan starts with gemini-2.0-flash. In the AI knowledge base, the model summarises tasks and chats, and the team uses semantic search available for Gemini.
The administrator switches the provider to Claude and pastes in a key from the Anthropic console; a warning then appears in Settings → AI that the new provider doesn't support embeddings, so semantic search stops working and only full-text search remains. They then refresh the list of models and select claude-opus-4-8. They also increase the maximum response length from 4096 to 8192 tokens.
Frequently asked questions
Which AI providers can I connect to Task Manager?
Task Manager supports OpenAI, Google's Gemini, Anthropic's Claude and xAI's Grok. Get an API key from your chosen provider and paste it into the AI settings.
Does response length affect costs?
Yes. You can set the maximum response length between 1024 and 16384 tokens, with a default of 4096. A higher limit allows longer responses, but increases the cost charged by the provider.
How do I change the provider or model?
In the AI settings, choose a provider, enter its API key, refresh the list of models and save your settings. The app tests the connection when you save. Semantic search works only with OpenAI and Gemini.
Is my API key secure?
The key is encrypted and never shown in full after you save it. Changing other AI settings won't overwrite it. You can enter a new key when changing the configuration.
What happens when I run out of provider credit?
Task Manager displays an error message in your language. When the provider rejects a request due to a limit, the application retries after 2 s, then after 8 s - up to 3 attempts in total. Only then does it show a message, and based on the provider's response it distinguishes an exhausted account budget from a temporary rate limit.