Web and Jev
builtin_fetch_url
HTTP GET of a public URL, returned as readable Markdown. Follows knoxchat.networkMode:
| Mode | Effect |
|---|---|
allow (default) | Public-host guard: private / loopback / link-local / CGNAT blocked unless KNOX_FETCH_URL_ALLOW_PRIVATE=1 |
deny | fetch_url blocked |
allowlist | Only knoxchat.networkAllowlist (exact or *.suffix). KNOX_NETWORK_ALLOWLIST comma-separated override |
Also: http/https only, no credentials in the URL, redirects re-validated (max 5 hops), time and size caps (2 MB body, 15 s, 60k output chars). Env: KNOX_NETWORK_MODE.
Known limit: DNS is checked before connect, so DNS rebinding between check and connect is still open. Data can still leak to a public URL if a query string carries it. Keep sensitive work in Ask/Edits and prefer deny / allowlist.
builtin_search_web
Deferred search. Results are untrusted text (same confused-deputy limits as files). Fetched / searched text is redacted on the way in.
Jev (System One)
knoxchat.jev.enabled (default on) asks a small harness for routing, skill hints, passage/tool gates, citation checks, and auto profile confirm.
| Selected models | Endpoint |
|---|---|
| KnoxStudio | https://api.knoxstudio.ai/v1/systemone |
| OpenRouter | https://openrouter.ai/api/v1/systemone (typesafe/jev-1.14) |
Sent: user message, short turn summaries, tool names and argument previews, truncated assistant text. Never file contents.

Each call has a latency budget (gates 800 ms, others 2 s) and a circuit breaker (3 failures → 5 minutes off). Fail-open: if Jev is down, chat still runs. YAML jev.enabled overrides the setting.
Status bar: knoxchat.jev.showStatusBar. Log: Knox: Show Jev Log. Off: knoxchat.jev.enabled: false.
Offline-ish
Model HTTP always goes to your provider. Combine networkMode: deny, Jev off, a local model, and disabled auto-update. There is no process-wide firewall. Full table: Network and offline.