Orylo Intelligence
Let your Assistants handle the little things and make Orylo work better for you.
Sections in this article
Set up Orylo Intelligence
Open Settings → Orylo Intelligence, select a feature, choose an Assistant, and enable it. Each feature has its own Assistant and optional Generation requirements for language, style, or emphasis. Leave requirements empty to use the built-in instructions.
Titles and tags are enabled by default, but still need a configured Assistant. The other features are off by default.
You can select a specific Assistant or follow the default Assistant. Open Assistant Settings to choose the model and its default reasoning effort; the available controls depend on the model. These are Assistant settings, so they also affect other uses of that Assistant. For separate settings, create an Assistant dedicated to Intelligence.
Conversation titles
After the first message you send, Orylo creates a short title so conversations are easier to find. A title you set manually takes priority. If generation fails, the existing title stays.
- Input: An excerpt from the first message you send; attachment files are not read.
- Output: A descriptive title of up to 60 characters, based on the task or topic you want help with.
- Assistant and reasoning: Choose a Chat Assistant for this feature; Agent Assistants are also supported. In the selected Assistant’s settings, use a lower Reasoning Effort, such as Low, if the model offers it. If it does not, keep the available defaults.
Summary previews
After a successful reply, a short summary becomes available when you hover over the conversation in the sidebar. It uses recent messages, so it is a quick reminder rather than a complete history. It does not replace memory or a fork handoff summary.
- Input: Excerpts from up to six recent messages, rather than the complete conversation.
- Output: A preview of up to three sentences and 600 characters covering the goal, key decisions, and unresolved questions.
- Assistant and reasoning: Choose a Chat Assistant for this feature; Agent Assistants are also supported. In the selected Assistant’s settings, use a lower Reasoning Effort, such as Low, if the model offers it. If it does not, keep the available defaults.
Fork handoff summary
Enable this feature, choose a branch point and target Assistant, then request a summary in the fork panel. Review and edit it before using it to hand off the conversation. The original local history remains available.
If the summary cannot be generated or the conversation exceeds the model’s limits, use the full-history option. Attachment names may appear in the summary, but the generator does not read the files.
- Input: Messages up to the chosen branch point and attachment names.
- Output: An editable summary of goals, constraints, decisions, progress, and remaining work, with references back to source messages.
- Assistant and reasoning: Choose a Chat Assistant for this feature; Agent Assistants are also supported. In the selected Assistant’s settings, use a lower Reasoning Effort, such as Low, if the model offers it. If it does not, keep the available defaults.
Memory suggestions
Enable memory as well as this feature. Orylo reviews new conversation excerpts during idle periods and proposes lasting preferences or agreements. Open Memory → Pending changes to review proposed additions, updates, or removals before accepting them.
Enabling this feature does not scan older conversations, and Team conversations are excluded from background organization. Suggestions do not silently overwrite your saved memories.
This feature drafts changes for review without tools. Asking an Agent to save a memory directly during chat is a separate operation that requires memory tools.
- Input: New conversation excerpts, relevant saved memories, and pending suggestions.
- Output: A proposed set of additions, updates, or removals for long-term memory. You decide whether to accept the changes.
- Assistant and reasoning: Choose a Chat Assistant for this feature; Agent Assistants are also supported. In the selected Assistant’s settings, use a lower Reasoning Effort, such as Low, if the model offers it. If it does not, keep the available defaults.
Action suggestions
Enable this feature to find repeated workflows across new conversations in the same workspace. It needs evidence from at least two independent conversations and runs after inactivity, at most once per day. Existing Actions are considered so suggestions can improve an Action rather than duplicate it.
Review and edit suggestions in this feature’s settings. Continue with Action Creator to open a draft in an Agent chat, then send it and confirm the final configuration. A suggestion never automatically creates or runs an Action.
Generating the suggestion needs no tools. Creating the Action with Action Creator is a separate step that requires a tool-capable Agent Assistant.
- Input: Excerpts from up to 12 new conversations in the same workspace and the existing Actions catalog.
- Output: An editable proposal to create or improve an Action, with supporting conversation references; it is not an executable Action.
- Assistant and reasoning: Choose a Chat Assistant for this feature; Agent Assistants are also supported. In the selected Assistant’s settings, use a lower Reasoning Effort, such as Low, if the model offers it. If it does not, keep the available defaults.
Next prompt suggestions
After a new successful reply, an eligible conversation with at least two completed replies may show a suggested next message in the empty input box. It follows your conversation language, including Chinese. Press Tab to turn it into an editable draft; it is not sent automatically. Typing takes priority.
Remote Web can display suggestions generated on your Mac; on mobile, tap the button to use the suggestion. Opening a conversation does not generate one. Suggestions can return when you reopen the same unchanged conversation during the current app session, but disappear when Orylo restarts.
- Input: Up to 12 recent completed messages in the current conversation, within a limited context budget.
- Output: One short next-message suggestion in the conversation’s language, or no suggestion when no useful next message is clear.
- Assistant and reasoning: Choose a Chat Assistant for this feature; Agent Assistants are also supported. In the selected Assistant’s settings, use a lower Reasoning Effort, such as Low, if the model offers it. If it does not, keep the available defaults.
Privacy and troubleshooting
These features use limited conversation excerpts, not attachment contents or private Agent history. A remote model receives the text required for the selected feature; an on-device Assistant processes it on your Mac. Normal provider usage charges may apply.
If nothing appears, check that the feature is enabled and its Assistant is available, then check its trigger conditions in the information panel. No useful suggestion, insufficient context, or a failed request can produce no result. Intelligence failures do not block normal chat.