What AI providers does Claw-ED work with?

Anthropic (Claude), OpenAI, Google Gemini, Ollama (local or cloud), and OpenRouter. You bring your own API key. The setup wizard helps you configure it.

Is my data sent anywhere?

Working data is stored locally. Hosted AI providers receive prompts and selected content. Enabled search, image, messaging, and Google integrations contact their services. Local inference requires a local model; network-dependent features still need a connection.

How does voice learning work?

When you run clawed ingest, it reads your lesson plans, worksheets, and assessments. It extracts vocabulary patterns, question framing style, scaffolding moves, and pedagogical structure. Generated content then uses those patterns.

Claw-ED — Your AI Teaching Assistant

Name: Claw-ED
Author: Claw-ED

Start with the lessons you already teach

Claw-ED helps a teacher turn existing curriculum into new lesson drafts, student materials, and slides. Bring your files, describe the class and learning goal, then review and edit the results in the tools you already use.

Bring your materials. Import PDF, DOCX, PPTX, text, or Markdown files. Claw-ED extracts content for search and builds a teaching-style profile.
Ask for a concrete lesson. Give the topic, grade, subject, and outputs you need. Your saved materials and profile can inform the draft.
Review and edit. Check sources, answer keys, pacing, accessibility, and layout before classroom use.
Choose how to deliver. Keep editable files locally or use integrations you have configured.

Try it with one lesson

The current beta starts from a terminal. A local browser dashboard is available after setup. You will need Python 3.11 or newer and access to a supported model.

python -m venv .venv
# macOS / Linux:
source .venv/bin/activate
# Windows PowerShell: .venv\Scripts\Activate.ps1
python -m pip install --upgrade clawed
clawed setup
clawed ingest ./my-lessons/
clawed lesson "Causes of the French Revolution" -g 10 -s "Global History"

Use clawed serve to start the local dashboard, or clawed for interactive use. Run clawed <command> --help for the options supported by that command.

Getting started · Install from PyPI · Troubleshooting

Useful outputs, with an honest status

The lesson bundle tool creates three core files from shared lesson content. Optional extensions can add other materials; a fixed file count is not guaranteed.

Core lesson bundle

Teacher plan · DOCX Teaching notes, timing, questions, and answer keys Student packet · DOCX Student-facing activities and handouts Lesson slides · PPTX An editable presentation built from the lesson content

Automated checks catch missing sections and some content problems. Failed checks trigger bounded repair attempts. Core bundle exports are checked individually; incomplete delivery is reported as partial or failed, and an unsuccessful package review is labeled draft. These checks do not replace teacher judgment.

Curriculum and style: search imported material and use a saved teaching-style profile. Inspect retrieval and source attribution before relying on them.
More teaching formats: assessment drafts, differentiation, games, simulations, and additional export formats, with feature-specific setup and limitations.
Connected work: Telegram, Google integrations, and MCP tools when configured. Verify permissions and the actual delivery destination.
Experimental classroom features: live sessions, student-facing tools, and scheduling. Some session state does not survive a restart.

Claw-ED is designed for one teacher per instance. It is not a multi-tenant school platform.

What changed in v9.18.2026.1

One owned Python runtime. No bundled third-party terminal or Node.js requirement.

Keep your selected model and isolate concurrent tool requests.
Queue, cancel, and resume lesson drafts with saved generation phases.
Preserve multilingual source text and review quotation provenance in a source manifest.
Use current local, budget, and premium model options.

Release notes and upgrade details

Choose the model for your budget

Checked September 18, 2026. These are starting recommendations; teacher review and your own comparison matter.

Use	Starting options	Tradeoff
Local laptop	Qwen 3.5 4B / 9B, Gemma 4 12B	No per-token fee; leave memory for context and other apps.
Larger local machine	GPT-OSS 20B, Qwen 3.8 27B	More memory and time; compare using your actual lesson task.
Budget cloud	OpenRouter: GPT-OSS 20B, Gemma 4 31B, Qwen 3.8 Flash	Low token prices; content goes to hosted providers.
Premium	GPT-6 Astra, Claude Fable 5.1	Explicit choices for complex work; higher token prices.

Exact model IDs, prices, memory estimates, setup, and official sources

Your provider sets access, pricing, and quotas. Claw-ED does not include usage. Ollama Cloud and free OpenRouter endpoints have limits.

Local storage, clear data boundaries

Configuration and working data are stored locally under ~/.eduagent/ by default. Hosted models receive prompts and selected content for the tasks you run. Web search, image retrieval, Telegram, Google integrations, and package installation contact their respective services when used.

For local inference, configure a local model and avoid cloud providers and network-dependent features. Review what you ingest and share, remove student identifiers where possible, and follow your school's data rules.

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

Read the security and privacy model.

Help improve the complete teaching workflow

The most useful contributions make import, drafting, review, and export work better together: reproducible bugs, source-fidelity checks, teacher-reviewed examples, and clearer output status.

CI checks Python 3.11 and 3.12, wheel installation, and Docker startup. Most model tests use synthetic data or mocks. Passing tests does not establish classroom quality across live models.

Contributing guide · Report an issue · Roadmap

Claw-ED is maintained by MacxLabs. Supporting development is optional.