Plan Mode: Think Before You Act + new models
Think different with Plan mode

Hey everyone,
We're thrilled to introduce Plan Mode to Alter, a new approach to complex workflows that puts thinking before execution.
Here's the simple idea: better planning leads to better results.
Plan Mode asks clarifying questions upfront, lets Alter understand your intent deeply, and creates a structured plan you review before anything happens.
Why Plan Mode Matters
When you're working on something complex, extracting data from multiple sources, building a workflow with specific requirements, or solving a problem with many moving parts, the difference between a rough request and a well-understood one is enormous.
Traditional mode works great for quick tasks. But for complex work, jumping straight to execution can mean:
Missed details that require rework
Assumptions about what you actually need
Changes made before you've had a chance to review the approach
Plan Mode inverts this flow. It makes Alter pause, ask questions, and think through the problem before taking action.

How It Works
Activate Plan Mode — Select Plan mode in the Hub, on the bottom left
Answer Clarifying Questions — Alter asks targeted questions to understand your intent
Review the Plan — You see a structured breakdown of the approach before execution begins
Approve & Execute — Only after you've reviewed does Alter move forward
Think of it as having an experienced collaborator who asks "before we do this, can you help me understand..."
New models
We’ve been adding new models in the past few weeks, Sonnet 4.6, GLM 5, Minimax 2.5, Qwen3.5-397B-A17B and FM2-24B-A2B-Preview to the roaster.
Voxstral 2 is also becoming our main STT cloud processor, keeping Deepgram Nova as a fallback.
We'd love your feedback. Hit us up in Discord or email if you have thoughts on how Plan Mode works for you and if you have a new favorite model in town.
See you next week.
Olivier & Samuel
Full Changelog
New Features & Enhancements
Plan Mode: Introduced Plan Mode to help structure complex workflows by planning before execution
Flow Engine: Add transform/chaining support between tools and a summarization step to avoid context bloat
New Models: Added Sonnet 4.6, GLM 5, Minimax 2.5, Qwen3.5-397B-A17B, and FM2-24B-A2B-Preview to the model roster
Speech-to-Text: Voxtral 2 is now the main cloud STT processor, with Deepgram Nova kept as a fallback
Bug Fixes & Stability
Hub Input & Navigation: Fixed hub input expansion and context strip navigation issues
Hub UI Stability: Fixed flickering scroll with tool calls and removed transparency effects from conversation view
Conversation & Sidebar: Fixed conversations sometimes not showing in sidebar, plus performance optimizations
Malformed Tool Calls: Server-side fixes to avoid malformed tool calls
Usage: Improved fair-use display behavior
Transcript Indexing: Fixed transcripts renamed from the notch not being reindexed properly
Menus & Action Editor: Fixed picky click hit areas in Alter menu categories and removed unwanted autofocus in action editor general pane
Microphone Sessions: Improved recording stop flow by clearing live sessions correctly, avoiding stray sessions, removing duplicate reset handling, improving stop reactivity, and storing failed recordings in history when possible
Microphone Errors: Better handling of session start failures
Floating Widget: Reverted change breaking context retrieval