Skip to main content

Longer AI Runs and Vibe Mode Manual

This release makes long-running agent work much more practical. Agents can stay inside one processing window longer, handle more tool activity, and keep multi-step execution more stable.

What changed​

The platform now does a better job in five areas:

  • longer AI processing windows
  • stronger sequential execution
  • better stop control
  • better context-window management
  • a more useful Vibe Mode

Together, these changes make long conversations and autonomous execution much more reliable.

Faster and more stable AI processing​

You should notice three visible improvements:

  • fewer cases of duplicate or double processing
  • fewer misleading timeout situations
  • faster overall processing in normal use

Internally, the runtime, retry budgets, and agentic loops were hardened so long executions stay inside safer operating limits.

Current expectation:

  • processing is roughly 50% faster in many common cases
  • more improvements are already in progress

Stronger sequential processing​

Long or data-heavy runs used to be more likely to skip steps or lose flow.

This release improves that behavior:

  • sequential processing is more consistent
  • multi-step runs are less likely to skip steps
  • agents handle tool-driven execution more reliably inside the same processing window

This matters most when a master is building on many documents, many prior messages, or multiple tool calls.

Agents can work longer in one run​

Agents now support longer runs with more tool activity inside a single processing cycle.

That means:

  • more tool calls can happen before the run completes
  • agentic loops are more reliable
  • long reasoning chains hold together better

This is one of the main reasons the platform now feels more capable on heavier sessions.

Vibe Mode: faster exploration, with a clear tradeoff​

Vibe Mode is now a more meaningful option for exploratory work.

Use it when you want the agent to keep moving across a short sequence of steps with minimal interruption.

Typical use:

  • up to 5 to 10 steps
  • fast alternative exploration
  • rough first-pass deliverables
  • momentum over precision

Important caveat​

Vibe Mode is still best for exploration, not final approval.

Why:

  • generated content may be slightly weaker in depth or nuance
  • some deliverables may need a human review pass before they become the final truth

Good use cases:

  • trying multiple solution directions
  • generating fast working drafts
  • exploring a workflow before committing to one path

Not ideal for:

  • final strategic approval
  • irreversible decisions
  • high-confidence executive artifacts without review

History limit and context window controls​

Next to the model selector, you now have a direct control for how much history the next turn sends.

Options:

  • All (default)
  • 20
  • 10
  • 5

This control limits how many recent messages and their related documents are sent into the next processing turn.

Why this matters​

If your session has many documents and many turns, you can now keep the conversation going longer without overloading the context window.

Use cases:

  • keep All when the full thread still matters
  • drop to 20, 10, or 5 when the current task only needs recent context
  • reduce context when performance starts to degrade in very large sessions

Context gauge​

The same control area now makes context usage easier to see.

You can:

  • see how much context is currently being consumed
  • reduce the next-turn payload
  • immediately understand the reduction after lowering the history limit

This gives you a practical way to trade context breadth for longer, cleaner conversations.

Shared stop control​

Longer runs need a stronger safety control.

This release improves stop behavior by making it:

  • more durable
  • more reliable in real time
  • safer for collaborative sessions

If multiple people are looking at the same session, stop control is now far more dependable than before.

Fast track​

  1. Start with All.
  2. If the session becomes very large, move to 20, 10, or 5.
  3. Use lower history settings when you only need recent context.
  4. Use Vibe Mode for speed and exploration, not final sign-off.
  5. Stop a run when needed without fearing that the shared state will drift as easily as before.

FAQ​

Does Vibe Mode replace normal guided work?
No. It is a fast-track mode for exploration.

Should I use All all the time?
No. All is the default, but shorter windows are useful for very large sessions.

Why would I lower the history limit?
To keep long conversations running well when the session already contains many documents and messages.

Is longer AI processing only about speed?
No. It is also about stability, fewer skipped steps, stronger loops, and more predictable control.