vs RTK / CodeGraph / Context Mode / QMD

This page is an honest audit. We list features both ways — wins and gaps with explicit roadmap items.

Quick verdict by use case

If your main use is…	Best fit
Rust + git + cargo + docker, AI agent context	ctx (single binary, all-in-one)
Pure command output compression, 100+ commands	RTK (deeper coverage, 53k stars)
Heavy call graph navigation in 19+ languages	CodeGraph (more languages, framework routing)
Session continuity across context compactions	Context Mode (unique feature)
Pure semantic search over docs, fully local LLM	QMD (more advanced search pipeline)

ctx wins when you want all of the above in one binary rather than orchestrating four separate tools.

Full feature matrix

	ctx	RTK	CodeGraph	Context Mode	QMD
Stars	—	53k	22k	15.5k	25.5k
Language	Rust	Rust	TypeScript	TypeScript	TypeScript
License	MIT	MIT	MIT	MIT	MIT
Hook PreToolUse auto-rewrite	✅	✅	❌	❌	❌
Native MCP server	✅ (10 tools)	❌	✅ (9 tools)	✅	✅
Command output compression	✅ (17 families)	✅ (100+ commands)	❌	✅	❌
Curated repo map	✅ (BM25 + PageRank)	❌	❌	❌	❌
Token budget binary search	✅	❌	❌	❌	❌
Local RAG over docs	✅	❌	❌	partial (FTS5)	✅
Call graph	✅ (7 langs)	❌	✅ (19+ langs)	❌	❌
Query-ranked graph results	✅	❌	❌	❌	❌
Token budget on graph output	✅	❌	❌	❌	❌
Session continuity (PreCompact)	🚧	❌	❌	✅	❌
Multi-agent installer	1 (Claude Code)	13	5	15	partial
Single binary	✅	✅	❌ Node bundled	❌ Node bundled	❌ Node bundled
100% local	✅	✅	✅	✅	✅

Where ctx wins clearly

1. Only tool with hook + MCP server + compression + RAG + repo map + call graph in one binary. Everyone else specializes vertically.
2. Only tool with query-ranked graph results. CodeGraph returns a raw list; we rank by BM25(query) + popularity + symbol kind.
3. Only tool with token budget on graph output. CodeGraph can overflow context with large traces; we apply binary search like ctx map.
4. Only tool with Personalized PageRank for repo map. No competitor does ranked code-aware selection.

Where ctx still loses

Vs RTK (command compression)

• 100+ commands vs 17 families — RTK covers ansible, sops, helm, gcloud, dotnet, swift-build, xcodebuild, mypy variants, prisma, basedpyright, etc.
• 13 agent installers vs 1 — RTK auto-configures Cursor, Copilot, Gemini, Windsurf, Cline, Hermes, etc.
• Telemetry / rtk gain — RTK has detailed savings dashboard.

Plan: add filters incrementally based on user demand. The pacote completo PR closed 70% of the gap (terraform, AWS by service, all major linters, log dedup, git ok-style).

Vs CodeGraph (call graph)

• 19+ languages vs 7 — missing C#, PHP, Swift, Kotlin, Scala, Dart, Svelte, Vue, Lua, Liquid, Pascal/Delphi, Luau.
• Dynamic dispatch / inheritance — interface → impl resolution.
• Framework-aware routing — Django/Rails/Express/Spring URL → handler.
• Live file watcher — FSEvents/inotify. We re-index on demand.

Plan: add languages on demand (each is ~80 LOC). Watcher is ~200 LOC. Framework routing is one strategy per framework.

Vs Context Mode (session continuity)

• PreCompact + SessionStart hooks — Context Mode persists session state across context compactions. Unique feature, no other competitor matches.

Plan: explicit roadmap item. ~300-500 LOC + design decisions on what state to capture.

Vs QMD (RAG quality)

• Reciprocal Rank Fusion (RRF) — QMD combines BM25 + vector results with RRF.
• Query expansion via LLM — QMD expands the query before searching.
• LLM reranker with logprobs — QMD uses the model’s confidence to rerank top results.

Plan: add RRF (~150 LOC), query expansion (~80 LOC), confidence-aware reranker.

Roadmap priority

Based on the gaps above and likely impact:

1. More agent installers — Cursor, Codex, opencode (trait already in place)
2. RRF in ctx catalog — cheap, broad recall improvement
3. Live file watcher for ctx graph — fixes the “stale index” annoyance
4. More languages in ctx graph — start with C#, PHP, Kotlin
5. Session continuity — ctx session save/restore
6. More ctx exec filters — helm, gcloud, ansible, prisma based on demand

See docs/competitors/ on GitHub for the raw per-competitor analysis.

Honest disclaimer

ctx is pre-1.0 (currently v0.1.0). The competitors are mature and battle-tested with thousands of stars and millions of downloads. Choose accordingly:

• For production use today, especially in Cursor/Copilot/Windsurf workflows, RTK or CodeGraph have more installers and broader command coverage.
• For Claude Code workflows that want everything integrated, ctx is competitive and getting closer every release.
• For experimenting with the future of agent context tooling, all five are worth knowing.