Context Mode helps AI coding agents use their context window intelligently. It keeps bulky tool output out of the conversation, preserves important session state across compaction, and encourages agents to solve data-heavy problems by writing code rather than reading everything into context.
Large tool results—browser snapshots, GitHub issue lists, logs, file reads, and web pages—can consume huge amounts of an agent’s context. Context Mode sandboxes that output, indexes what matters, and retrieves only the relevant pieces when needed, helping longer coding sessions stay focused and useful.
Features
Context-window optimization — Sandboxes raw tool output so large results do not flood the agent’s context; the project reports up to a 98% reduction in stored context for supported workflows.
Session continuity — Tracks file edits, Git operations, tasks, errors, and user decisions in SQLite, then retrieves relevant history through FTS5 full-text search and BM25 ranking.
“Think in code” workflow — Encourages the agent to write a script that performs analysis and returns only the result, instead of repeatedly reading files or running broad commands.
Sandboxed execution tools — Includes tools for batch execution, JavaScript execution, file-based execution, indexing, search, and fetch-and-index workflows.
MCP server integration — Works through the Model Context Protocol, with 11 ctx_* tools for context saving, retrieval, diagnostics, upgrades, and statistics.
Automatic routing — On hook-capable platforms, hooks steer the agent toward context-efficient tools and capture session events without requiring you to rewrite every prompt.
Broad platform support — Provides setup paths for Claude Code, Gemini CLI, VS Code Copilot, JetBrains Copilot, GitHub Copilot CLI, Cursor, OpenCode, KiloCode, Codex CLI, Kimi Code, Qwen Code, Kiro, Zed, Antigravity, and OpenClaw/Pi Agent.
Diagnostics and visibility — Run ctx doctor to validate runtimes, hooks, FTS5, and plugin registration; run ctx stats to inspect token use, per-tool savings, and efficiency.
Optional local knowledge base — Index files or directories for later semantic-style retrieval, with commands to search, upgrade, or permanently purge indexed content.
Open source — Distributed under an open-source license and installable globally via npm or through platform-specific plugin marketplaces.
Croco Loki added Context Mode as alternative to Caveman, Headroom and RTK
No comments or reviews, maybe you want to be first?