Features

Feature Overview

简体中文 · Home

Models and coding operations

GrayCode supports Gemini, OpenAI Compatible, OpenAI Responses, and Anthropic. The assistant can read, search, and edit files, create directories, run terminal commands, and query symbols, definitions, and references through VS Code language services. File changes are normally presented as native diffs, and sensitive tools can require confirmation.

The input box accepts files, folders, selected code, images, audio, video, PDFs, and other context. Supported channels send attachments in native multimodal formats instead of treating binary data as text.

Structured engineering workflow

• Design records requirements, constraints, interfaces, solutions, and risks under .graycode/design/.
• Plan turns an approved design into steps and TODOs under .graycode/plans/.
• Progress maintains stages, risks, milestones, and next actions in .graycode/progress.md.
• Review stores evidence, findings, and conclusions under .graycode/review/.

These artifacts keep design, implementation, and review traceable across long-running or collaborative work.

Conversations and long-running tasks

• Multiple chat tabs preserve separate work contexts.
• Messages can be queued while the assistant is busy or inserted at tool boundaries.
• Background commands and Sub-Agents return results to the main conversation when they finish.
• Automatic summarization compresses older context near configured limits.
• Tool state, reasoning, tokens, caching, TTFT, and response latency remain visible.

Branching conversations and checkpoints

Rerolling or editing a message can create a candidate branch instead of overwriting the old response. Branches can be switched, renamed, soft-deleted, and restored. When a branch performed workspace writes, you can optionally restore its linked workspace checkpoint as well.

GrayCode can create checkpoints around important messages and tool batches, and the input toolbar supports manual checkpoints. Restoration checks for unsaved files before changing the workspace.

Streaming and visualization

Streaming supports adjustable smoothing, character fades, and incremental Markdown rendering. A TPS indicator visualizes generation speed, and Mermaid blocks render as diagrams.

The usage page groups tokens, cache activity, and estimated cost by conversation, model, and date. IDE activity tracking stores timestamps only, not conversation content, and provides daily, monthly, and schedule heatmap views.

Extensible and local-first

MCP connects external tools, Skills provide on-demand knowledge, and Sub-Agents handle specialized tasks. OptMem keeps conventions and decisions in separate global and workspace scopes. Core settings, history, checkpoints, and memories remain local.

Continue with Tools and Workflows or Extensions and Memory.