Added an adaptive “Thinking Deeper” mode to improve planning and overall quality of larger edits or changes.

Fixed rate limit (429) error backoff to be enforced server-side, not just in the JS.

Added web search capability via DuckDuckGo HTML result parsing so the agent can access the web natively.

Changed all system tools to function like an MCP server so they’re easier to expand on later.

Fixed an issue where the Markdown response from the AI occasionally ended up in a table if the markdown was processed before the streaming finished.

Added subagent executions for parallel workflows and file restrictions for each agent which can be assigned by the main agent orchestrator to make sure they only touch what they need to, then are destroyed.

Added a date function. Why not?

Feature requests, anyone?