Upcoming Platform Improvements
We're working on several improvements to make your experience with ISMS Copilot faster, more cost-effective, and easier to use—especially in long conversations.
We know the current long conversation setup isn't good enough. Our #1 engineering priority right now is implementing message compaction to enable much longer conversations without severely impacting your usage limits. While longer conversations will always consume more tokens due to how AI works, compaction will bring you much closer to infinite conversations with minimal usage impact. Expected delivery: late February to early March 2026.
Message Compaction (Top Priority - Late Feb/Early March 2026)
Long chat threads currently resend the entire conversation history with each message, which quickly consumes your token quota and causes "conversation too long" errors. We're implementing message compaction with automatic context management that will:
Enable near-infinite conversation length without hitting token limits
Reduce usage impact by up to 90% for ongoing conversations through smart summarization
Preserve critical context (uploaded files, audit findings, control numbers, decisions) while compacting older messages
Maintain recent messages in full for immediate context
The same conversation that might use 50,000 tokens today will use only 5,000-10,000 tokens after compaction
This is our key focus and will ship between late February and early March 2026.
Additional Long Conversation Improvements
Along with compaction, we're adding:
In-chat warnings when a conversation is using significant credits, with smart suggestions on when compaction will help vs. when to start fresh
Clearer token usage information so you can see exactly how much each message costs and why
Visual indicators showing conversation health and compaction status
Smarter document detection
Every message currently triggers a separate check to see if you're requesting document analysis. We're optimizing this process to skip unnecessary checks and reduce background token usage you never see.
Fairer billing model
We're reviewing whether to charge only for new content (your latest message and the AI's response) rather than the full conversation history that gets resent with each message. This would make long conversations far more affordable.
These improvements are in development. We'll update this article and notify users as features roll out.
When will these ship?
Prompt caching and document detection optimization are our highest priorities and will ship first. Context management and billing model changes require more testing but are actively in progress.
Have questions or suggestions? Contact our support team—we're listening.