Imported from HKUDS/DeepTutor (AGENTS.md). Install upstream with npx skills add HKUDS/DeepTutor. Copyright stays with the author.

DeepTutor — Agent-Native Architecture

Overview

DeepTutor is an agent-native intelligent learning companion organized around a two-layer plugin model — single-shot Tools invoked by the LLM, and multi-stage Capabilities that take over a turn — exposed through three entry points: CLI, WebSocket API, and Python SDK.

Architecture

Entry Points:  CLI (Typer)  |  WebSocket /ws  |  Python SDK
                    ↓                   ↓                   ↓
              ┌─────────────────────────────────────────────────┐
              │              ChatOrchestrator                    │
              │   routes UnifiedContext → selected Capability    │
              │   (defaults to `chat`)                           │
              └──────────┬──────────────┬───────────────────────┘
                         │              │
              ┌──────────▼──┐  ┌────────▼──────────┐
              │ ToolRegistry │  │ CapabilityRegistry │
              │  (Level 1)   │  │   (Level 2)        │
              └──────────────┘  └────────────────────┘

All capabilities emit on a shared StreamBus; the orchestrator fans events out to consumers. Runtime settings live in data/user/settings/*.json — project-root .env files are intentionally ignored.

Level 1 — Tools

Single-function tools the LLM picks on demand. Four user-toggleable tools surface in /settings/tools:

The rest are context-gated: the chat capability auto-mounts them from ToolMountFlags (presence of a KB, attachments, sandbox availability, …), and any of them can also be force-enabled via --tool. Auto-mounted set: rag, read_source, read_memory, write_memory, read_skill, load_tools, exec (sandboxed Python/C/C++ or shell execution), list_notebook, write_note, web_fetch, github, cron, ask_user (pauses the turn and resumes with the user's reply), plus the mastery-path tools. geogebra_analysis is parked under COMING_SOON_TOOL_TYPES.

Level 2 — Capabilities

Multi-stage pipelines that own the turn:

All capabilities converge on emit_capability_result() in deeptutor/capabilities/_shared.py so every turn emits the same envelope (response payload + cost_summary from UsageTracker). Status copy and prompts are i18n'd via capabilities/prompts/{en,zh}/<name>.yaml.

CLI Usage

# Install
pip install deeptutor      # Full app (CLI + Web/API + packaged Web assets)
pip install deeptutor-cli  # CLI-only

# Run any capability
deeptutor run chat "Explain Fourier transform"
deeptutor run deep_solve "Solve x^2=4" -t rag --kb my-kb
deeptutor run visualize "Animate sine wave" --config render_mode=manim_video

# Interactive REPL
deeptutor chat
# (inside the REPL: /regenerate or /retry re-runs the last user message)

# Partners (IM-connected companions)
deeptutor partner list

# Knowledge bases, memory, server
deeptutor kb list
deeptutor kb create my-kb --doc textbook.pdf
deeptutor memory show
deeptutor serve --port 8001       # API server only
deeptutor start                   # backend + frontend together

Key Files

Dependency Layers

Public install paths and source extras are defined in pyproject.toml. Requirements files mirror the same dependency groups for Docker/CI installs.

pip install deeptutor      — Full app (CLI + Web/API + packaged Web assets)
pip install deeptutor-cli  — CLI-only (LLM + RAG + providers + document parsing)
pip install -e .           — Source install for development

Source extras (.[ extra ], defined in pyproject.toml):
.[cli]            — CLI-only dependency set
.[server]         — Web/API server dependencies
.[partners]       — Partner channel SDKs  (legacy alias: .[tutorbot])
.[matrix]         — Matrix channel for Partners (matrix-nio; needs libolm)
.[matrix-e2e]     — Matrix with end-to-end encryption (matrix-nio[e2e])
.[math-animator]  — Manim addon (powers `visualize` Manim renders + `deeptutor run math_animator`)
.[dev]            — Test / lint tooling
.[all]            — Everything above