A coding agent that runs on your machine: search, edit, run, and verify, all through the local codepapr-server host. The main agent coordinates Explore / Scout / Mentor, with byte-level optimization for the LLM prefix cache.
Ask · Plan · Agent · App — one runtime, from explanation to execution to .papr apps
Explain architecture, trace responsibilities, analyze error causes. Read-only by default — no file changes or commands.
Read-Only · ExplainBefore big changes, lay out a plan, a task list, and how you'll verify it; ask when something is ambiguous, then execute once you confirm.
Plan · DecomposeSearches the project, reads and writes files, runs commands, verifies — then fixes bugs or ships features in one go.
Execute · Read/WriteGenerate interactive .papr apps (charts, boards, panels), or install from the official market into global or project scope — isolated origin / CSP / SQLite.
Visualization · Market InstallFile Tree · Monaco · Git · Browser · LSP · Plugin Market — all in one interface
Full project browsing, Monaco highlighting, LSP-first (definition/references/hover/call hierarchy), with automatic AST project-graph fallback when LSP is unavailable.
Type / for commands. Keyboard navigation.
Search, file edits, commands, and verification — all automatic. The agent loop runs in @codepapr/core on the desktop; file and command work goes through the local host so the UI stays responsive.
Diff, commit, branch switch, safe rollback. Agent handles analysis; human gates high-risk operations.
Launch preview, page interactions, screenshots. Frontend debugging in one window.
Auto-create checklist for complex tasks, track progress in real-time, auto-collapse on completion. Auto-retry on failure.
/goal: main Agent runs, Verifier checks, until the verification condition passes.
Auto-commit before each user message. Reset --hard from any message; code and chat roll back together.
Right-side round bar. Click to jump.
Preferences, constraints, and verified facts are curated into .CodePapr/MEMORY.md — readable, editable, diffable. The panel is the editor; saves take effect next turn. How context is stacked.
Install .papr from the official registry into global or project scope; isolated origin, CSP, and private SQLite, with app_publish push.
The main agent hands work to Explore (code), Scout (web), and Mentor (architecture) via task; Verifier, Compactor, and the memory curator run behind the scenes.
Desktop and codepapr-cli share the same local codepapr-server (JSON-RPC). The capability lives in codepapr-core; clients only call into it. Tutorial
Create or import cards (PNG/JSON). Talk in that voice when enabled. Off by default — turn it on under Settings → General → Experimental features.
GPT-SoVITS-based voice synthesis from reference audio. Streaming WebSocket output with multilingual support. Off by default — turn it on under Settings → General → Experimental features.
All projects below were independently designed, coded, and organized by CodePapr
Claymore anime fan curation site — character profiles, plot guides, community
View Project →Dallas Mavericks NBA fan hub — team history, player stats, season tracking
View Project →The Boys fan curation site — character profiles, episode breakdown, world-building
View Project →New York championship memorial site — title runs, key games, glory moments
View Project →Stock index display case — for learning only, not financial advice
View Project →20-day coastline cycling route — real map, daily itinerary and scenery
View Project →
/ to Invoke19 built-in commands, auto-routed to primary or fast model by task complexity; /help /commands /undo bypass models, zero token cost
| Command | Purpose | Model |
|---|---|---|
/help /commands | List all available commands and descriptions | Local |
/compact | Force-compress current session context — deterministic skeleton checkpoint (zero LLM; one second-level summary only if the skeleton still overflows) | Primary / Fast |
/undo | Undo the last conversation reset (restore truncated messages and code snapshot) | Local |
/goal | Autonomous loop: main Agent executes + Verifier audits, until the verification condition passes | Primary |
/review | Review changes or specified scope, prioritize bugs and regression risks | Primary |
/fix | Locate and fix issues, auto-run verification | Primary |
/test | Add tests for specified feature or run existing tests | Primary |
/explain | Explain file, function, symbol, or error implementation approach | Primary |
/diagnose | Diagnose errors, slow operations, or abnormal behavior root causes | Primary |
/refactor | Safe refactoring, preserving behavior | Primary |
/doc | Write or update documentation (.md extension only) | Primary |
/new | Create new file, component, or feature from description | Primary |
/optimize | Analyze and fix performance bottlenecks, reduce complexity or resource usage | Primary |
/build | Build project and diagnose/fix build errors | Primary |
/search | Search codebase for patterns, usage, definitions, or references | Fast |
/lint | Run linter and fix violations | Fast |
/clean | Clean dead code, unused imports, and leftover debug statements | Fast |
/commit | Stage changes and generate standardized commit message | Fast |
/summary | High-level overview of file, module, or entire project | Fast |
.CodePapr/commands/<name>.md or Project Config with $ARGUMENTS, @path file embedding, and !`cmd` inline outputs. Read Tutorial →
File I/O, commands, and Git go through the codepapr-server host; the agent loop and prompts live in the TypeScript client. Tauri is the shell, not the agent.
Desktop · CLI
Same JSON-RPC
Local host daemon / 159+ RPC · newline-delimited JSON-RPC
Agent loop · cache · tools / Explore / Scout / Mentor
Sessions · checkpoints · plugin UI / workspace source of truth
Tauri · Monaco · browser / .papr · Stronghold · TTS
To extend or optimize, read Host–Client architecture and context layering.