Build with the AI agents

Work with the chat agents on one map and across all your maps.

Open the chat and pick an agent#

Press Map chat in the toolbar, or press C if the map is yours. Who answers depends on the picture you are in: on the canvas it is the Builder, in the graph view it is the Graph analyst (it finds connections you have not made). On the canvas, click the agent's name at the top of the chat to switch: Reader answers questions about what the map says or where it is weak, and changes nothing.

Writer writes the titles, subtitles and bodies of blocks already on the map; it adds none. Orchestrator sends work to the map's branches and reports back. Only the map's owner can use it, and only in the map's own conversation.

Each branch can have a conversation of its own. Click the conversation's name at the top of the chat (it starts as "This map"), then pick a block, or search for it by name once the map has many. That conversation changes and adds blocks only inside its branch. Its Reads switch decides what it can see: Branch loads only that branch, Whole map also reads the rest, for context. Switching conversation keeps the mode, except Orchestrator, which works only in the map's own conversation; switching mode keeps the conversation.

Ask across all your maps#

On the dashboard, the button in the corner (Ask across every map you have) opens Across your maps. Ask where something is, or where a new note belongs: "Which of my maps talk about pricing?" It finds and suggests; it never edits a map.

A map's own chat answers from that map only. To let it quote your other maps, open the chat's menu (Chat and map options), then Search, then All my maps.

Plan, build and review#

When a request is too big for one go, the Builder writes a plan of up to 12 steps and works through them, keeping its place between your messages. A bar in the chat shows where it is ("step 2 of 5"); Show the steps unfolds it and Discard the plan drops it.

Self-review has no button: the agent checks what it built and writes a review of the connections, the layout and the colours, with a drawing of the flaws when its model can read pictures.

Teach it your rules#

A skill is a standing rule the agent follows while it builds. Open Settings ▸ Skills & Scenarios, press New skill, write a title and the instructions, and choose where it applies (Every map or Chosen maps) and which agents follow it. To switch one on for a single map, use ⋮ ▸ Map settings… ▸ Skills or the chat's Skills menu. A skill an assistant writes for you over MCP starts on no map until you choose its maps here, so the assistant cannot put a rule in force on its own. A scenario is an interview the agent leads: see Let the map interview you.

Speak instead of typing#

Press the microphone (Dictate a message) in the chat box, talk, and press it again to finish; Esc discards. Your words land in the message box to edit and send. The microphone appears only where dictation is available, and has its own allowance: see your plan.

Reasoning level and your own key#

Settings ▸ AI Chat Model picks the model that builds your maps. A model marked Included in your plan needs no setup. Any other runs on your own key: find its provider (DeepSeek, Anthropic, OpenAI and Gemini), paste the key and press Save. The provider then bills you directly and Corlecti Map does not meter the run. What the included models use is on the AI allowance.

Reasoning effort sets how long the model thinks. Its levels depend on the model, and a model with none offers none. A single map can differ: in its chat menu, Effort takes a level for that map only, or Use account setting. A Telegram bot has its own model and level on its card, and an MCP key gets its pair when you create it. A bot set to "A chat turn" runs on your own key when you have saved one for its provider, and its turns then count against your MCP calls, not the AI allowance. Choose Corlecti Map's shared key on the card and they count against the allowance.