Changelog

Agent skills, self-improving agents, and model tiers

S
S
M
T
W
T
F

This week we’re giving your digital coworkers better memory, the ability to learn new procedures on the fly, and a way to improve their own performance based on how you interact with them.

Agent skills

You can now attach reusable modules of knowledge and procedures to any agent. Instead of writing long, repetitive instructions for every agent, you can define skills once and let your agents load them when they need them.

  • Procedural playbooks — Define "how we file a bug" or "our brand voice" as a skill shared across your team.
  • On-demand activation — Agents only load the full text of a skill when it becomes relevant, keeping their conversations focused and efficient.
  • Workspace authored — Create your own skills in Settings → Skills to capture your team's unique domain knowledge.

Self-improving agents

Your agents now learn from your feedback. When you give a thumbs-up or thumbs-down to an agent's response, Monad analyzes the gap between the actual response and what you expected.

  • Instruction refinement — The system periodically optimizes an agent's internal instructions to improve its accuracy.
  • Safety first — Every proposed improvement is measured against a history of your past feedback to ensure there are no regressions before it goes live.
  • Transparency — You can review, approve, or revert any auto-generated instruction changes in the Agent Optimization dashboard.

Agents now dream

To keep your long-term conversations coherent, agents now perform a nightly memory consolidation pass. We call this "dreaming."

  • Coherent history — Agents deduplicate, merge, and prune their stored memories overnight, resolving contradictions and tidying up stale facts.
  • Contextual relevance — By tidying their own notes while you sleep, agents stay focused on what matters most to your team during the day.

Virtual model tiers

We’ve introduced three virtual model tiers—Frontier, Balanced, and Cost-effective—to give you more control over the balance between intelligence and cost.

  • Frontier — The most capable models for complex reasoning and creative tasks.
  • Balanced — A blend of speed and intelligence for your everyday workflows.
  • Cost-effective — Fast, lightweight models for high-volume tasks like basic summarization.

Improvements

  • Admin identity — Team admins can now access a new platform portal with support for federated identity (OIDC).
  • Onboarding polish — A smoother setup experience with better message tracking and clearer navigation.
  • Legal suite — Completely updated Privacy, Terms, and Data Processing policies grounded in our commitment to transparency.
  • Refined feedback — Better UI for rating agent responses and seeing the impact on their instructions.

Fixes

  • Browser reliability — Fixed an authentication loop that affected some Safari users.
  • Integration reliability — Resolved an issue where some third-party tools would fail to load on high-performance models.
  • UI Polish — Fixed sidebar flickering and tooltip overlapping issues during heavy message activity.