Claw-ED

Claw-ED

Your curriculum. Your teaching voice. Lessons you can review and edit.

Claw-ED is a local-first AI teaching assistant. Start with your existing lesson plans, slides, and assessments; use them to draft new lessons, student materials, and presentations. You choose the model and review the result before using or sharing it.

Current release: v9.18.2026.1 · Beta · Python 3.11+

Install · Website · Release notes · Security and privacy · Report an issue

Start with one lesson

1. Bring your materials. Import a folder of PDF, DOCX, PPTX, TXT, or Markdown files. Claw-ED extracts text, indexes material for retrieval, and builds a teaching-style profile.
2. Give it a concrete task. Specify the topic, class, learning objective, and materials you need. Existing material and your saved profile can inform the draft.
3. Review the outputs. Check source accuracy, answer keys, pacing, accessibility, and layout. Edit the DOCX and PPTX files in your usual tools.
4. Choose what to deliver. Keep files locally or use configured integrations. Student access and external publication need deliberate setup.

The lesson bundle tool creates a teacher DOCX, student DOCX, and PPTX slides. Optional extensions can add differentiated materials, games, and other formats. The actual result depends on the request, model, installed extras, and successful exports; a fixed file count is not guaranteed.

Setup

Use Python 3.11 or newer in a virtual environment:

python -m venv .venv
# macOS / Linux:
source .venv/bin/activate
# Windows PowerShell: .venv\Scripts\Activate.ps1
python -m pip install --upgrade clawed
clawed setup

Choose a provider and configure access in the setup wizard. Then try the Python commands:

clawed ingest ./my-lessons/
clawed lesson "Causes of the French Revolution" -g 10 -s "Global History"
clawed serve                         # local dashboard and API
clawed --version

Run clawed for the owned Python interface; Node.js and Bun are no longer required. clawed -p "request" uses the same gateway and tool policy. For command-specific options, use clawed <command> --help.

Getting started · Troubleshooting

What is available today

Area	What it provides	What to check
Curriculum ingestion	Text extraction, search, teaching-style profiles, and image extraction	Retrieval may miss or shorten relevant material; inspect the sources used
Lesson production	Structured lesson content, teacher/student documents, and slides	Factual accuracy, source attribution, answer keys, and classroom fit
Quality checks	Section, source-text, differentiation, and assessment checks with bounded repair attempts	These are automated checks, not proof that a lesson is ready to teach
Dashboard and API	Local lesson views, teacher chat, exports, and student embed snippets	Teacher authentication and correct share-link setup
Integrations	Telegram, Google Drive, Classroom-related tools, and MCP	Provider credentials, permissions, and destination readback
Experimental features	Games, simulations, scheduling, live classroom sessions, and student-facing tools	Feature-specific limitations; some session state does not survive restarts

The release is intended for a single teacher per instance. It is not a multi-tenant school platform. Student-facing features remain experimental within Claw-ED.

What v9.18.2026.1 changes

• Remove the bundled third-party terminal runtime; all entry points use Python.
• Keep your chosen model, isolate concurrent tool requests, and support GPT-6 Astra and Claude Fable 5.1 request formats.
• Add a Jobs page and CLI cancel, recover, and resume commands with saved generation phases.
• Preserve multilingual text and full source excerpts; include a source manifest that flags quotations needing review.
• Refresh local and budget model recommendations against current provider catalogs.

For a recoverable bundle, use the dashboard’s Jobs page or:

clawed queue submit bundle --topic "River civilizations" --grade 9 --subject History
clawed queue worker
# In another terminal, when needed:
clawed queue cancel JOB_ID
clawed queue recover
clawed queue resume JOB_ID

The worker must be running to process jobs. Completed matching phases survive restarts; external publishing and arbitrary chat sessions are not automatically resumed. Existing configuration and teaching data are preserved. See the changelog.

Models and privacy

Provider adapters support Anthropic, OpenAI, Google Gemini, Ollama, and OpenRouter. Model support and tool behavior vary by provider and task. API pricing, quotas, and model availability are set by the provider; Claw-ED does not include model usage.

Start here: dated model recommendations and setup. Try Qwen 3.5 9B locally, GPT-OSS 20B or Gemma 4 31B on OpenRouter for a budget trial, and explicitly select Astra or Fable for premium work. Recommendations are starting points, not measured classroom rankings.

Configuration and working data are stored locally under ~/.eduagent/ by default. Local storage does not mean all processing stays on the device. Hosted models receive the prompts and selected content needed for a request. Web search, image retrieval, Telegram, Google integrations, and package installation also contact their respective services when used.

For local inference, configure a locally running model and avoid cloud providers and network-dependent features. Review what you ingest and share; remove student identifiers where possible. The application is not a certification of compliance with school data policies.

Teacher web routes require authentication. The dashboard uses an HttpOnly session cookie with same-origin checks on changes; API clients can use a bearer token. Student embeds use lesson share tokens and separate conversation tokens. Keep the server local unless you have configured access controls for the intended audience.

Read SECURITY.md for the trust boundary and reporting process.

Useful commands

clawed                                  # interactive assistant
clawed ingest ./my-lessons/              # import teaching materials
clawed lesson "Topic" -g 8 -s "History"  # draft a lesson
clawed unit "Topic" -g 9 -w 3            # draft a multi-week unit
clawed assess "Topic" --type crq         # draft an assessment
clawed kb query "question"               # search the curriculum wiki
clawed bot                              # Telegram interface
clawed serve                            # local dashboard and API
clawed drive auth                       # configure Google Drive access
clawed schedule list                    # inspect configured schedules
clawed mcp-server                       # expose tools to an MCP client

Development

git clone https://github.com/SirhanMacx/Claw-ED.git
cd Claw-ED
python -m pip install -e ".[dev]"
ruff check .
mypy --strict clawed
pytest tests/

CI exercises Python 3.11 and 3.12, wheel installation, and Docker startup. Most tests use synthetic data or mocked model responses; passing CI does not establish teaching quality across live models. See the architecture and evaluation protocol.

Useful contributions include reproducible bugs, teacher-reviewed sample lessons, source-fidelity checks, and improvements to a complete import → draft → review → export workflow. See CONTRIBUTING.md and the roadmap.

Claw-ED’s original code is released under the MIT license. Third-party components retain their own terms. The project is maintained by MacxLabs; supporting development is optional.