Text improvement that stays grounded in your workspace
Highlight any copy in Vibe Manager to clean it up with the configured text-improvement models. The rewrite preserves formatting, records token usage, and drops straight back into the Monaco editors, task descriptions, or terminal dictation buffers you were already using.
Popover-driven workflow
The text improvement provider listens for selections in Monaco editors and standard inputs. When text is highlighted it opens a floating Sparkles button beside the cursor, queues a background job with the selected range, and swaps in the improved content once the job completes.
- Works in the plan viewer, merge instructions, and plan terminal input
- Skips replacement if you edit the selection while a job is running
- Background jobs sidebar shows the original and improved snippets
Model configuration you control
Improvement jobs call the text-improvement task type defined in the AI settings. By default the workspace uses Claude Sonnet 4 with Gemini 2.5 Flash as the fallback, a 4,096 token ceiling, and a temperature of 0.7, but you can change the allowed models or overrides in configuration files.
- System prompt keeps the original language and formatting intact
- Non-streaming LLM requests expose prompt and completion token totals
- Project-aware settings come from the same resolver as other AI jobs
Voice dictation stays editable
Voice transcripts land directly in your task description or terminal buffer with project-specific language defaults. You can highlight the dictated text immediately and run the same improvement button without leaving the flow.
- Project-level transcription settings include model, language, and temperature
- Audio level monitoring warns about silence before you waste a take
- Improvement payloads keep a reference to the source transcription job
Video analysis feeds better prompts
Screen recordings bundle your current task description with optional attention prompts and pass them to the Gemini video analysis pipeline. The resulting notes appear alongside text improvement jobs so you can refine summaries before generating plans.
- Frame-rate slider keeps uploads within the model’s token limits
- Optional microphone capture adds transcripts to the analysis output
- Cost estimates come back with every video run
Guardrails built into the job system
Every improvement request runs through the same job queue and error handling that powers the rest of Vibe Manager. If authentication, network, or provider issues crop up, the UI surfaces the failure and leaves the original text untouched.
- Automatic retries respect the configured max concurrency
- Token usage and system prompt template are stored with the job
- Improved text is only applied when the response is complete
Experience Contextual Text Enhancement
From selection to perfection, workspace-aware improvements at your fingertips. This is how text improvement should work - grounded, integrated, effortless.