Batch 21 — CC GUIs, Session Viewers & Native Desktop Companions
Roster (10)
| slug | stars | distribution | cli_binary | local_ui | orchestration | multi_model | tier |
|---|---|---|---|---|---|---|---|
| opcode | 21,948 | desktop-app (Tauri 2) | none | desktop (React + Rust) | none | no | A |
| claudia-gui | 21,948* | desktop-app (alias) | none | desktop | none | no | C (alias → opcode) |
| poirot | 184 | desktop-app (Swift) | none | desktop (SwiftUI) | none | no | A |
| anima | 15 | desktop-app (Tauri v2) | none | desktop (Vanilla JS + Rust) | none | no | A |
| agent-fm | 40 | desktop-app (Electron) | none | menu-bar | none | yes (Gemini/OAI for narration) | A |
| cchub | unknown | unknown | unknown | unknown | none | unknown | C (repo not found) |
| mantra | 7 | desktop-app (Rust+React) | none | desktop | none | no | A |
| patapim | unknown | desktop-app | none | desktop | none | no | C (closed source, website only) |
| vibe-island | 54 | desktop-app (Swift) | none | menu-bar/notch | none | no | A |
| aiderdesk | 1,233 | desktop-app (Electron) | none | desktop (React+TS) | hierarchical | yes (30+ providers) | A |
*claudia-gui star count inherited from opcode (same repo)
Intra-batch patterns
All 10 frameworks in this batch add no prompting methodology — they are visual/audio shells around existing AI coding CLIs, not workflow frameworks. The shared pattern is: read AI coding session data (JSONL or live streams) and present it through a richer interface. Five tools are macOS-only (Anima, Agent FM, Vibe Island, PATAPIM, and Poirot); only AiderDesk and Mantra are cross-platform. The batch splits along two temporal axes: live monitoring (Agent FM, Vibe Island, Anima — watch agents as they run) vs. post-session forensics (Poirot, Mantra — review what happened). AiderDesk is unique in being a full agentic orchestration platform rather than a companion viewer; it wraps Aider CLI (not Claude Code) and has the most sophisticated architecture in the batch.
Most interesting find
Agent FM stands out for the most sophisticated observer prompt in the corpus — LIVE_NARRATOR_PROMPT defines a complete multi-agent narration protocol with bracketed input taxonomy, prioritization rules, and hard output constraints. It is the only tool in the entire Phase D corpus that delivers primary output as audio speech, and it intentionally uses Gemini/OpenAI to narrate Claude/Codex agents (cross-vendor observer pattern).
AiderDesk stands out for having the most powerful extension API in the batch — 30+ lifecycle hooks, runtime TypeScript tool injection, and injectable React UI components. It ships its own .aider-desk/ Claude Code skills/agents/commands for its own development (meta-application of methodology), making it the most self-referential tool in the batch.
Items written as Tier C
claudia-gui — Identified as an alias/rename of
opcode. The "Claudia" branding was the original name of the opcode project before renaming. Full analysis is in theopcodeentry.canonical: falsein METRICS.yaml.cchub —
https://github.com/Moresl/cchubreturns null from GitHub API (repo not found, private, or deleted). No public material available.patapim — Closed-source commercial product with minified website. No GitHub repository found. Technical analysis limited to website meta description. Key finding: uses browser automation to route to Claude.ai subscription instead of API — legally ambiguous but architecturally unique in corpus.
Cross-references discovered
- opcode / claudia-gui: Same project. opcode was originally named "Claudia" (all forks still carry that name).
getAsterisk/opcode=winfunc/opcode=basilisk001/claudia= same codebase. - Poirot and superpowers: Poirot's
AI_INSTRUCTIONS.mdreferences skills by name (sf-symbols-expert,swiftui-expert-skill,swift-testing-expert,swift-concurrency) — it was developed using a skill-based Claude Code workflow, making it an implicit consumer of skill-based methodology from the seed corpus. - AiderDesk and spec-kit (seed): AiderDesk's SKILL.md pattern (frontmatter + progressive disclosure) matches spec-kit's SKILL.md format exactly, suggesting adoption of the same skill specification format.
- AiderDesk and kiro (seed): Both use worktree isolation per task and emphasize human approval before code lands. AiderDesk is the open-source equivalent of kiro's control model.