0.36.0 Beta 1

Dyad v0.36.0 Beta 1 is focused on making it easier to start complex work with more clarity and fewer “agent hiccups.” This release introduces Planning mode, improves the robustness and reliability of Agent mode, and adds a new OpenRouter model option: Moonshot AI’s Kimi K2.5.

Highlights

Planning mode

This beta introduces Planning mode—a new workflow that helps you turn a vague request into a clear plan before AI starts making changes.

Planning mode is designed for situations like:

building a new app with complex requirements
starting a new feature where you want to align on approach first
multi-step tasks where you don’t want to “discover the plan” mid-way through execution

In practice, Planning mode guides you through a short planning flow (including a lightweight questionnaire when needed), then produces a structured plan you can iterate on before moving into implementation.

Beta note

Planning mode is new and we’re actively tuning it. If the questions feel off (too many, too few, or missing key info), please share feedback—especially with a chat session upload.

More robust Agent mode

Agent workflows are now more resilient—especially in longer sessions and when dealing with large repos.

Key improvements:

Better handling of long contexts: Dyad now includes a context compaction system for local-agent mode, helping reduce failures when sessions get large.
Safer context gathering: grep-based searches are now constrained to avoid runaway results that can trigger context-length errors.
Ask mode is more reliable: it now always uses the local agent handler, reducing mismatches between the UI flow and agent execution.
Fewer “stale reference” errors.

Feedback appreciated

If you run into agent failures that feel like “it worked yesterday” (context errors, stuck sessions, or odd history issues), please upload a chat session so we can reproduce it quickly.

New model option: Moonshot AI’s Kimi K2.5 (via OpenRouter)

Dyad now supports Moonshot AI’s Kimi K2.5 as an OpenRouter model option. Kimi K2.5 is a state-of-the-art open-source model and an excellent value option.

Enhancements & fixes

Added a native notification setting when a chat stream completes (useful when you’re multitasking).
Added a fuzzy search to the Settings sidebar, making it much faster to find the setting you need.
Added a chat panel visibility toggle, so you can show/hide the chat panel depending on your workflow and screen space.
Chat input now supports recalling previous messages with the Up Arrow, making it easier to resend or tweak prompts.
Improved chat auto-scroll reliability by replacing fragile timeout-based tracking with position-based logic.
Added optional line range parameters to the read_file tool, making it easier to focus the agent on just the relevant portion of a file.
Fixed an issue where creating a new chat could incorrectly assume the main branch; Dyad now uses your current branch.
Improved error recovery for Pro users with a “Start new chat” button directly in the error state.

Feedback

This is a beta release, so feedback is especially helpful.

If you run into issues or have suggestions (especially around Planning mode, Agent reliability in long sessions), please help us improve by reporting a bug or uploading a chat session.