AgentWeb Next is a safer, smaller rebuild of the original AgentWeb idea: web access for AI agents with structured search, fetch, and research outputs.
This repo fixes the main issues found in thienduy2211/agentweb:
- Modular architecture instead of giant
core.py/deep_research.pyfiles. - Tests are first-class and must pass in CI.
- Docs and install paths point to
thienduy2211/agentweb-next. - Localhost/private IP fetches are blocked by default to reduce SSRF risk.
- No heavy reranker/model download in the default path.
- No undocumented auth/stealth promises. Experimental features must be exposed before being documented.
uv tool install 'agentweb-next @ git+https://github.com/thienduy2211/agentweb-next.git'Or for development:
git clone https://github.com/thienduy2211/agentweb-next.git
cd agentweb-next
uv sync --extra dev
uv run pytestagentweb-next search "Hermes Agent" --format json
agentweb-next fetch https://example.com --format json
agentweb-next research "Python agents web access" --format markdownfrom agentweb_next import AgentWeb
aw = AgentWeb()
print(aw.search("python agents"))
print(aw.fetch("https://example.com"))
print(aw.research("python agents"))fetch blocks private, loopback, link-local, unspecified, multicast, and cloud metadata IPs by default.
For local development tests only, pass an explicit policy in Python:
from agentweb_next.safety import URLPolicyCLI intentionally does not expose an allow-localhost flag yet.
This is an initial clean MVP, not a drop-in clone of the old repo. It prioritizes correctness, safety, packaging, tests, and maintainability.