Web Search

Search the web in a headless and fast way.

Search Workflow

1. Run the codex exec search
2. Open or fetch the most relevant pages
3. Merge, deduplicate, and use the findings accordingly
4. Repeat if needed

Codex exec Search

Use codex exec non-interactively and capture only the final assistant message:

tmp="$(mktemp)"
err="$(mktemp)"
trap 'rm -f "$tmp" "$err"' EXIT

if codex -m gpt-5.3-codex-spark exec --skip-git-repo-check --ephemeral \
  --output-last-message "$tmp" \
  "Search the web for: <QUERY>" \
  >/dev/null 2>"$err"
then
  cat "$tmp"
else
  cat "$err" >&2
  exit 1
fi

• Do not use --json here. It emits the full event stream and pollutes context
• --output-last-message writes the final assistant message to a file
• Codex still writes run metadata and logs to stderr
• Redirect stderr to a temp file and replay it only on failure
• Use -C <DIR> when search needs repo context

Example:

tmp="$(mktemp)"
err="$(mktemp)"
trap 'rm -f "$tmp" "$err"' EXIT

if codex -m gpt-5.3-codex-spark exec --skip-git-repo-check --ephemeral \
  --output-last-message "$tmp" \
  "Search the web for the latest stable Rust release and cite the best sources" \
  >/dev/null 2>"$err"
then
  cat "$tmp"
else
  cat "$err" >&2
  exit 1
fi

Readable page fetch

Use r.jina.ai to fetch readable text or markdown for specific pages.

curl -fsSL "https://r.jina.ai/https://example.com/article"

Notes:

• Prepend https://r.jina.ai/ to any HTTP or HTTPS URL
• Output is readable text or markdown for the target page

Default Behavior

When asked to search the web:

• run codex exec search
• fetch the most relevant pages as needed
• use r.jina.ai when a readable page view helps
• provide an answer with the best results