Research Skills
Research Skills give the OpenClaw Agent the ability to "search the web." From basic web browsing to AI-powered intelligent search, these Skills let the Agent access the latest information in real time rather than relying solely on training data.
#2 — Web Browsing
| Property | Details |
|---|---|
| Rank | #2 / 50 |
| Category | Research |
| Total Score | 70 / 80 |
| Maturity | 🟢 Stable |
| Official/Community | Official (bundled) |
| Installation | Bundled — no installation needed |
| ClawHub Stats | 180K+ usage count |
| Target Users | All users |
Feature Overview
Web Browsing is one of OpenClaw's most essential Skills, built directly into the system:
- Web navigation: Visit any URL, parse HTML content
- Content extraction: Intelligently extract article body text, tables, and code blocks
- Search integration: Query via search engines (Google by default)
- JavaScript rendering: Handle SPAs and dynamically loaded content
- Screenshot capability: Capture visual snapshots of web pages
- Multi-tab management: Browse multiple pages simultaneously
Why It Matters
Web Browsing is the key Skill that transforms the Agent from a "closed system" into an "open system." Without it, the Agent can only rely on training data; with it, the Agent can access real-time web information. The 180K+ usage count speaks to its indispensability.
Score Breakdown
| REL | COM | TRC | VAL | MNT | RLB | SEC | LRN | Total |
|---|---|---|---|---|---|---|---|---|
| 10 | 10 | 10 | 9 | 8 | 7 | 7 | 9 | 70 |
Ranking rationale: Perfect scores in relevance, compatibility, and community traction. Reliability is slightly lower (some websites block automated access), and security requires attention (the Agent could be tricked into visiting malicious sites).
Configuration
# Web Browsing is a bundled Skill — ready to use
openclaw skill status web-browsing
# Custom settings
openclaw config set web-browsing.default_search google
openclaw config set web-browsing.timeout 30000
openclaw config set web-browsing.javascript true
# Set proxy (often needed on corporate networks)
openclaw config set web-browsing.proxy http://proxy.corp.com:8080
Dependencies & Security
- Dependencies: Chromium runtime (installed with OpenClaw)
- Permissions Required: Network access
- Security: SEC 7/10 — the Agent may be lured to malicious sites via prompt injection
Malicious websites may embed prompt injection attacks targeting AI Agents. Recommendations:
- Set a URL whitelist:
openclaw config set web-browsing.allowed_domains "*.github.com,*.stackoverflow.com" - Enable sandbox mode:
openclaw config set web-browsing.sandbox true - Avoid letting the Agent automatically click unknown links
- Alternatives: Tavily (#4) provides more structured search results; Firecrawl (#18) is better for large-scale crawling
#4 — Tavily
| Property | Details |
|---|---|
| Rank | #4 / 50 |
| Category | Research |
| Total Score | 67 / 80 |
| Maturity | 🟢 Stable |
| Official/Community | Community (framix-team) |
| Installation | clawhub install framix-team/openclaw-tavily |
| Target Users | Researchers, users needing high-quality search |
Feature Overview
Tavily is a search engine designed specifically for AI Agents:
- AI Search: Semantic understanding of query intent, returning structured results
- Web crawling: Deep-crawl specified URLs and extract structured content
- Instant answers: Directly answer factual questions (with source citations)
- Search depth control:
basic(fast) vsadvanced(deep) - Source credibility scoring: Each search result includes a credibility score
Why It Matters
Compared to Web Browsing's general browsing capability, Tavily focuses exclusively on "search" and excels at it. Its results are pre-processed by AI, so the Agent can use them directly without additional HTML parsing. For research-intensive workflows, Tavily is more efficient than Web Browsing.
Score Breakdown
| REL | COM | TRC | VAL | MNT | RLB | SEC | LRN | Total |
|---|---|---|---|---|---|---|---|---|
| 9 | 9 | 8 | 9 | 8 | 8 | 8 | 8 | 67 |
Installation & Setup
clawhub install framix-team/openclaw-tavily
# Set the Tavily API Key (free plan: 1000 queries/month)
openclaw skill configure tavily \
--api-key tvly-xxxxxxxxxxxx
# Usage example
openclaw run "Use Tavily to search for the latest OpenClaw version features"
Dependencies & Security
- Dependencies: Tavily API Key (free plan available)
- Permissions Required: Network access (Tavily API only)
- Security: SEC 8/10 — only accesses the Tavily API, no direct web browsing, reducing injection risk
- Alternatives: Felo Search (#15) offers optimized Chinese-language search; Web Browsing (#2) is more general-purpose
#15 — Felo Search
| Property | Details |
|---|---|
| Rank | #15 / 50 |
| Category | Research |
| Total Score | 59 / 80 |
| Maturity | 🟡 Beta |
| Official/Community | Community |
| Installation | clawhub install felo-search |
| Target Users | Users needing multilingual search, Chinese-language users |
Feature Overview
Felo Search is another AI-powered search tool with distinctive features:
- Multilingual optimization: Chinese and Japanese search quality superior to Tavily
- Citation markup: Every answer sentence includes source links
- Instant answer mode: Generates summaries instead of link lists
- Topic tracking: Continuously track the latest information on specific topics
Why It Matters
For Chinese-language users, Felo Search typically outperforms Tavily in Chinese content search quality. If your primary research language is Chinese or Japanese, Felo Search is the better choice.
Score Breakdown
| REL | COM | TRC | VAL | MNT | RLB | SEC | LRN | Total |
|---|---|---|---|---|---|---|---|---|
| 8 | 7 | 7 | 8 | 7 | 7 | 8 | 7 | 59 |
Installation & Setup
clawhub install felo-search
# Set the API Key
openclaw skill configure felo-search \
--api-key your_felo_api_key
# Usage example (Chinese search)
openclaw run "Use Felo to search for the latest AI startup news in Taiwan"
Dependencies & Security
- Dependencies: Felo Search API Key
- Permissions Required: Network access (Felo API only)
- Security: SEC 8/10 — similar security model to Tavily
- Alternatives: Tavily (#4) for better English search quality; Web Browsing (#2) for the most general-purpose solution
#19 — Summarize (Research Perspective)
For the full introduction to the Summarize Skill, see Productivity Skills. In research workflows, Summarize plays the following role:
Summarize in Research Workflows
# Search → Summarize pipeline
openclaw run "
1. Use Tavily to search for 'transformer architecture 2026 improvements'
2. Summarize the top 5 results
3. Consolidate into a research note and save to Obsidian
"
The Summarize Skill works best in combination with Tavily or Web Browsing, compressing large amounts of search results into actionable insights.
Research Skills Comparison
| Feature | Web Browsing | Tavily | Felo Search | Summarize |
|---|---|---|---|---|
| Search capability | Yes | Yes | Yes | No |
| Web browsing | Yes | No | No | No |
| Crawling capability | Yes | Yes | No | No |
| Structured output | No | Yes | Yes | Yes |
| Chinese optimization | No | Average | Yes | Yes |
| Citation markup | No | Yes | Yes | No |
| Requires API Key | No | Yes | Yes | No |
| Works offline | No | No | No | Yes |
Recommended Research Combinations
# Deep research workflow
clawhub install framix-team/openclaw-tavily
clawhub install community/summarize
clawhub install community/obsidian-claw
# Paired with bundled Web Browsing
# Chinese research workflow
clawhub install felo-search
clawhub install community/summarize
clawhub install community/notion-claw
# Minimal research workflow (zero cost)
# Use only bundled Web Browsing + Summarize
clawhub install community/summarize