SafestClaw Assistant
Automate everyday tasks locally without an LLM, and add AI when a task calls for it.
- Source repo
- princezuda/safestclaw
- Stars
- ★ 282
- Last updated
- 1mo ago
- License
- MIT
- Primary language
- Python
- FA score
- 49/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformOpenAI API · Claude API
- Cost
- Free software; you pay for model usage
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A developer who wants a digest of enabled RSS sources can run
news techfor technology headlines. - Not a fit if
- Users who need autonomous multi-step reasoning or browser automation
- Users who require free-form AI conversation
- Source review
- 49/100 · Major gaps
What does this agent do, and when should you use it?
SafestClaw is a self-hosted assistant for Linux, macOS, and Windows, primarily used through its CLI. Its parser routes natural-language input to actions using keywords, regex, fuzzy matching, and date parsing; optional LLMs can handle blogging, coding, research analysis, and command understanding. It works with RSS feeds, web pages, files, email, and calendars, producing summaries, reminders, analysis, and blog drafts. Default features can run locally, while weather and services the user chooses to connect require external access; SafestClaw also offers a loopback-only Web UI, Telegram and webhook channels, and an optional FastMCP server. It suits users who want rule-based, auditable workflows with fewer API costs, but not those who need free-form chat, autonomous multi-step reasoning, or browser automation.
SafestClaw’s command parser routes input to actions using keywords, regex patterns, fuzzy matching, specificity-weighted phrase matching, and date parsing. Commands include news to read enabled RSS feeds and summarize them; summarize and crawl to extract web content; calendar to read ICS files or sync CalDAV; document to extract text from PDF, DOCX, HTML, Markdown, or TXT files; security scan to run installed scanners; and research to search arXiv, Semantic Scholar, or Wolfram Alpha. Results can be returned in the terminal or saved as summaries, reminders, blog text, or confirmed blog publications. Optional Whisper and Piper handle speech locally; with an LLM provider configured, SafestClaw can also write with AI, assist with coding, analyze research, or translate unrecognized input into a command.
- A developer who wants a digest of enabled RSS sources can run
news techfor technology headlines. - A researcher can search arXiv or Semantic Scholar, select sources, then decide whether to request LLM analysis.
- An engineer maintaining a project can run
security scanon an allowed path to check it with installed scanners. - Someone managing a personal schedule can read an ICS calendar or configure CalDAV sync and reminders.
- A writer can create deterministic blog drafts and titles without an LLM, then configure WordPress, Joomla, or SFTP publishing.
- A self-hosting user who prefers voice input can install Whisper and Piper for local speech-to-text and text-to-speech.
How do you install or deploy this agent?
Python 3.11 or newer is required. Install the core package with pip; features such as MCP, CalDAV, and Telegram require their optional extras. After launching, use the setup wizard to choose local features or configure an LLM, Web UI, MCP, or Telegram. Weather, online research sources, cloud LLMs, and remote publishing require network access; particular services may also require credentials.\n\n``bash\npip install safestclaw\npython -m safestclaw\npython -m safestclaw setup\n`\n\nInstall optional features as needed:\n\n`bash\npip install \"safestclaw[mcp]\"\npip install \"safestclaw[caldav]\"\npip install \"safestclaw[telegram]\"\n``
How do you use this agent?
Start the interactive CLI and enter commands. Calendar files, email, publishing targets, and external services need their respective setup before use. AI writing and deep research analysis require an LLM provider; deterministic blogging, summarization, and many local actions do not.\n\n``bash\nsafestclaw\n`\n\nExamples:\n\n`text\nnews tech\nsummarize https://example.com/article\ncalendar today\nresearch arxiv quantum computing\nsecurity scan ~/projects/myapp\n`\n\nOptionally start the loopback-only Web UI or expose tools as an MCP subprocess:\n\n`bash\nsafestclaw web\nsafestclaw mcp\n``
What are this agent's strengths and limitations?
- Core command parsing and many actions work without an LLM; deterministic blogging, summarization, and coding utilities can avoid paid models.
- Runs on Linux, macOS, and Windows, with CLI, loopback-only Web UI, Telegram, webhook, and FastMCP access modes.
- Combines concrete workflows for RSS, crawling, email, calendars, document extraction, code security scanning, and blog publishing.
- Supports multiple local or cloud LLM providers, with separate provider routing for blogging, research, and coding.
- The README says SafestClaw does not provide free-form chat, autonomous multi-step tasks, browser automation, or self-writing skills.
- CalDAV, Telegram, MCP, smart home, and vision features require extra dependencies or configuration; vision adds about 2GB.
- Weather, online research, social content, and remote publishing need network access; some services also need accounts, API keys, or target configuration.
- The README says default features work offline except weather, but cloud LLM use and unrecognized input sent through enabled NLU leave the machine.
How does this agent compare with similar options?
The README compares SafestClaw with OpenClaw. SafestClaw emphasizes rule-based commands and local features that can work without an LLM, and lists voice, RSS, research, and security scanning among its capabilities. It also states that it lacks free-form chat, autonomous multi-step reasoning, browser automation, and self-writing skills. Compare the two against the workflows you need rather than assuming they offer equivalent capabilities.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| SafestClaw Assistant This agent | 49 · Major gaps | CLIFree + model costs | ★ 282 | 1mo ago | Python | OpenAI API · Claude API |
| HexStrike AI MCP | 22 · Major gaps | MCP serverFree | ★ 13k | 3mo ago | Python | Claude.ai |
| Jev Browser | 75 · Some gaps | MCP serverFree + model costs | ★ 128 | 4d ago | JavaScript | Claude Code |
| Obscura Headless Browser | 75 · Good | CLIFree | ★ 30k | today | Rust | Claude.ai |
How does FollowAgents rate this agent?
Why each dimension lost points
The README describes local processing, optional LLM use, disclosure of NLU data sent to a provider, loopback-only web binding, scanner path limits, and blog previews. These provide partial data-flow and external-effect detail. Permission boundaries, protection for sensitive email/device data, confirmation and rollback flows, dependency risk handling, and source attribution are not sufficiently documented, so most criteria score 1.
The CI workflow shows lint and multi-platform test configuration, and the supplied tests cover parts of the LLM cascade and error classification. The README also describes offline notices and fallbacks. This does not establish runtime outcomes, and broad capability claims have limited supporting evidence; self-consistency and dependency availability remain thin. Failure messaging is described more concretely and scores 2.
The materials cover CLI, web, Telegram, MCP, offline use, and optional AI/ML, and state some clear limits, including no autonomous multi-step tasks or browser automation. The conditions and boundaries for natural-language input and semantic routing are less clear, so trigger precision scores 1.
The README is organized by feature and includes installation methods, example commands, and limitations. pyproject.toml supplies package name, version, dependencies, and a complete MIT license. The supplied README excerpt is truncated near its examples; FAQ, change history, and maintenance responsibility are limited, so those criteria do not receive high scores.
The CLI, research, summarization, reminders, and integrations can produce directly usable outputs, and the README describes rule-based local features that do not require an LLM. This indicates practical value and potential cost savings. However, feature and comparison claims rely mainly on the README, with little independent corroboration or detailed evidence about cost and output quality, so each scores 2.
The CI configuration and test source provide traceable evidence for some capabilities, but many README claims about features, security, privacy, and competitors have no corresponding implementation or independent source in the supplied files. The materials do not consistently distinguish implemented facts from promotional claims. No execution or independent verification is claimed.
- Many privacy, security, feature-coverage, and OpenClaw comparison statements in the README are project claims that the supplied materials cannot independently verify. Before using email, publishing, device-control, or external API features, review their configuration, permissions, and confirmation behavior.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
4.2.5 Securitty update: issue unlikely to impact most users
Release notes excerpt
This is a security update if you're using webhooks to grab third party content you haven't reviewed to avoid certain malicious attacks.View on GitHub ↗ -
4.2.4 Fixed edge cases causing LLM to not work
-
4.2.3 Updated to node.js twenty four