Planning now supports explicit interactive steps. Instead of forcing every decision into free-form chat, agents can pause on a structured interaction, wait for you, and then continue execution cleanly.
CHOICE Mode: Button-Based Decisions
When a step truly depends on your decision (not your prose), CHOICE makes that explicit.
- A clear question - The step presents a single question and waits for your input
- A bounded set of options (1-10) - Options show up as buttons so you can answer in one click
- Web UI buttons - While the run is
RUNNING, you will see choice buttons directly in the entity detail UI - Completion is recorded - After you choose, the run completes and the selected option is shown as the outcome
CHOICE In Telegram
If you have Telegram linked, CHOICE prompts can also be delivered as an inline keyboard.
- Inline buttons for each option
- Agent auto-switch so the prompt and reply route to the correct agent context
CALL_TO_ACTION Mode: Links + "Done"
Some steps are not about choosing, they are about doing something outside Kindship (opening a dashboard, completing a setup, approving access, etc). CALL_TO_ACTION makes those moments crisp.
- A direct link - The step includes a button that opens the relevant page or tool outside Kindship
- Optional button label - Customize the link button text (defaults to "Open Link")
- Web UI support - You will see an "Open Link" button and a "Done" button while the run is
RUNNING - Explicit completion - Clicking "Done" marks the step complete so orchestration can continue
CALL_TO_ACTION In Telegram
- Open Link button plus a Done confirmation button
- Agent auto-switch so replies and follow-ups stay coherent
Cross-Channel Completion (Idempotent)
Interactive steps complete consistently whether you respond on the web or in Telegram.
- Idempotent completion - If you already completed the step in another channel (for example, Telegram first), the web UI can safely acknowledge that and move on
- Account-safe responses - Your response is applied to the correct plan and agent context
More Reliable Tool Output (Less Chat Weirdness)
This release also includes several changes that reduce tool-related breakage:
- Tool result schema fixes - Tool result message parts are handled more consistently
- Anthropic prompt adjacency fixes - Tool results are less likely to be mis-ordered or merged incorrectly
- Fewer Telegram retries and tool-call spam - Telegram webhook retry behavior was tightened to avoid duplicate tool-call noise