Skip to main content

Additional Reference Projects

Archived comparison document. Preserved for historical context; not part of the current normative reference/ contract.

Extension of 16-reference-comparison.md. Candidates surfaced via WebSearch on 2026-04-12 across four categories:

  1. Monitoring/orchestration CLIs wrapping coding-agent CLIs
  2. UI wrappers over agent harnesses
  3. Multi-harness SDKs/runners
  4. Session/token analytics

Candidates

#NameURLStars (approx)LanguageCategoryAgents supportedRelevance to agent-mux
1cctophttps://github.com/st0012/cctop~1kSwiftMonitoringclaude-code, opencodeMenubar switcher between sessions — UX pattern for our sessions surface.
2ccmonitorhttps://github.com/shinagaki/ccmonitor~500TS/NodeMonitoringclaude-code5-hour rolling window reporting we don't model in CostRecord.
3cc-monitor-rshttps://github.com/ZhangHanDong/cc-monitor-rs~300Rust (Makepad)Monitoringclaude-codeNative real-time UI consuming JSONL — validates our event stream shape.
4Claude-Code-Usage-Monitorhttps://github.com/Maciek-roboblog/Claude-Code-Usage-Monitor~6kPythonAnalyticsclaude-codeML-based burn-rate prediction — gap in agent-mux.
5ccboardhttps://github.com/FlorianBruniaux/ccboard~400RustMonitoring + Dashboardclaude-codeUnifies sessions, costs, hooks, agents, MCP in one binary — mirrors our CLI surface.
6cc-viewerhttps://github.com/weiesky/cc-viewer~200TSMonitoringclaude-codeLive request/response capture for debugging — could inspire a --trace mode.
7claude-code-monitor (zcquant)https://github.com/zcquant/claude-code-monitor~150TSAnalyticsclaude-codeOpenTelemetry OTLP dashboard — telemetry export gap.
8ccusagehttps://github.com/ryoppippi/ccusage~20kTSAnalyticsclaude, codex, opencode, pi, ampJSONL session reader family — already tracked; listed for completeness.
9claude-usage (phuryn)https://github.com/phuryn/claude-usage~200TSAnalyticsclaude-codePro/Max subscription progress-bar UX.
10vibe-kanbanhttps://github.com/BloopAI/vibe-kanban~15kRustUI wrapperclaude, copilot, cursor, codex, gemini, qwen, droid, opencode, amp, ccrKanban + per-agent worktree — gap: we lack a default UI.
11papercliphttps://github.com/paperclipai/paperclip~30kTS/ReactSDK + UIclaude, codex, cursor, openclaw, pi, bash, httpAlready tracked; closest structural peer.
12claude-squadhttps://github.com/smtg-ai/claude-squad~7kGoTerminal orchestratorclaude, codex, gemini, aider, opencode, amptmux-based parallel workspaces with yolo mode — gap.
13aiderhttps://github.com/Aider-AI/aider~42kPythonAgent harnessmulti-LLM (Claude, o1, DeepSeek, local)Tree-sitter repo map + architect mode — we don't ship these.
14continuehttps://github.com/continuedev/continue~26kTSIDE wrapperany (Claude, GPT, local, Mistral)VS Code/JetBrains integration — agent-mux has no IDE surface.
15goose (block)https://github.com/block/goose~29kRustAgent framework15+ providers via MCP3000+ MCP tools + recipes — our MCP manager is narrower.
17OpenHarness (HKUDS)https://github.com/HKUDS/OpenHarness~400PythonHarnessClaude, OpenAI, Copilot, Codex, Kimi, GLM, MiniMaxAcademic harness with pluggable endpoints — reference for capability surface.

Cross-reference: features agent-mux lacks

FeatureCovered byPriority
Burn-rate / limit prediction (5-hour rolling)Claude-Code-Usage-Monitor, ccmonitorHigh — pairs with existing CostRecord.
OpenTelemetry / OTLP exportclaude-code-monitor (zcquant), ccboardMedium — we have events but no OTLP exporter.
Built-in Kanban / task UIvibe-kanban, paperclipMedium — scope decision (SDK-first vs UI-first).
Tree-sitter repo map / architect planningaiderMedium — out of scope for pure harness, but useful as plugin.
IDE extension (VS Code / JetBrains)continueLow — different shape of product.
MCP recipe library (3000+ tools)gooseMedium — we ship MCP lifecycle but no curated catalog.
Live request/response trace viewercc-viewerLow — debugging aid.
Menubar / system tray switchercctopLow — presentation layer only.
Qwen / droid / amp adaptersvibe-kanban, ccusage, claude-squadMedium — already noted in 16-reference-comparison.md.

Net assessment

Adding these 17 projects to the survey in 16-reference-comparison.md does not change the top-line conclusion: agent-mux remains a structural superset on the SDK surface. However three previously-unflagged gaps emerge:

  1. Predictive burn-rate analytics (Claude-Code-Usage-Monitor) — additive to CostRecord.
  2. OTLP telemetry exporter (zcquant, ccboard) — natural plugin.
  3. MCP recipe catalog (goose) — complements our MCP lifecycle manager.

The UI-shaped projects (vibe-kanban, paperclip, continue) remain orthogonal: agent-mux is an SDK, not a UI, and should stay that way per scope doc.