Examples & recipes

omniscout daemon start                              # 1. start the daemon
omniscout browser navigate https://example.com      # 2. drive the browser
omniscout browser snapshot --refs-only              # 3. find @eN refs
omniscout browser click '@e3'                       # 4. act on them
omniscout browser screenshot --out /tmp/out.png     # 5. capture state
omniscout browser close --all                       # 6. clean up

Use OmniScout to build a knowledge graph for "Cursor". Run
`omniscout graph "Cursor" --json --data`, then present the tree and list the
top sources. If the user gives a website instead, use site-only mode
(`omniscout graph "cursor.com"` or `omniscout graph "Cursor" -w cursor.com`).

OMNISCOUT_JSON=1 omniscout graph "Cursor" --data > /tmp/graph.json
jq -r '.entity, .nodes' /tmp/graph.json

# Site-only (no DuckDuckGo)
omniscout graph "https://cursor.com" --no-llm

Use OmniScout to research "what's the current state of open-source
browser-using AI agents in 2026?" Use `omniscout research` with 10 results
and JSON output, then summarize the top 3 passages with their source
URLs in a markdown table.

OMNISCOUT_JSON=1 omniscout research \
  "current state of open-source browser-using AI agents in 2026" \
  --results 10 > /tmp/r.json

jq -r '.passages[:3][] | "| \(.source_url) | \(.text[:120])... |"' /tmp/r.json

Open https://docs.python.org/3/library/asyncio.html via OmniScout, extract
the readable content with `omniscout extract`, then answer: "what's the
difference between asyncio.run and asyncio.gather?" Cite specific
sentences from the page in your answer.

URL=https://docs.python.org/3/library/asyncio.html
omniscout extract "$URL" --format markdown > /tmp/page.md
# Now your agent reads /tmp/page.md and answers.

Use OmniScout to remember https://docs.python.org and
https://docs.anthropic.com, then search your memory for "taskgroup" and
summarize what you find. Use --json throughout.

omniscout remember https://docs.python.org
omniscout remember https://docs.anthropic.com --note "API reference"
OMNISCOUT_JSON=1 omniscout search "taskgroup" --source memory
omniscout memory stats
omniscout workflow export --from-context

Use OmniScout to open https://duckduckgo.com, fill the search box with
"local-first AI agents", press Enter, wait for results to settle, then
list the title and URL of the top 5 results.

omniscout browser navigate https://duckduckgo.com
omniscout browser snapshot --refs-only > /tmp/snap.json
SEARCH_REF=$(jq -r '.refs[] | select(.role == "searchbox" or .name | test("search")) | .ref' /tmp/snap.json | head -1)
omniscout browser fill "$SEARCH_REF" "local-first AI agents"
omniscout browser key Enter
omniscout browser wait --idle
omniscout browser snapshot --refs-only \
  | jq -r '.refs[] | select(.role == "link") | "\(.name)"' | head -5

Help me set up a logged-in GitHub profile via OmniScout:
1. `omniscout profile create work`
2. `omniscout browser login https://github.com/login --profile work`
3. Pause for me to authenticate in the browser window.
4. Once I say "done", run `omniscout browser navigate
   https://github.com/notifications --profile work --session work` and
   take a screenshot to /tmp/notifs.png.

omniscout profile create work

# Terminal A: opens browser, blocks
omniscout browser login https://github.com/login --profile work

# Terminal B (after you log in in the browser):
omniscout browser login --done

# Now reuse:
omniscout browser navigate https://github.com/notifications \
  --profile work --session work
omniscout browser screenshot --session work --out /tmp/notifs.png

Open https://example-with-captcha.com via OmniScout. If `omniscout browser
captcha --detect-only` reports a CAPTCHA, run `omniscout browser captcha`
without a solver — that will flip the tab headful and block until I
solve it. Tell me what to do, wait for the block to return, then
continue with the task.

Open https://example-with-captcha.com via OmniScout. If there's a CAPTCHA,
solve it via 2Captcha — TWOCAPTCHA_API_KEY is set. Use `omniscout browser
captcha --solver 2captcha`. Continue only after `ok: true`.

omniscout browser navigate https://example-with-captcha.com
DET=$(omniscout browser captcha --detect-only)
if echo "$DET" | jq -e '.detected' >/dev/null; then
  omniscout browser captcha --solver 2captcha  # opt-in
fi

Use OmniScout to investigate which trackers Vercel's pricing page loads:
1. Start network capture.
2. Navigate to https://vercel.com/pricing.
3. Scroll to the bottom.
4. Stop capture.
5. List any requests whose URL matches "stripe|analytics|datadog|posthog".
6. For each match, run `network detail` and tell me the HTTP status.

SES=pricing-audit
omniscout browser network start --session $SES
omniscout browser navigate https://vercel.com/pricing --session $SES
omniscout browser scroll down --amount 10 --session $SES
omniscout browser network stop --session $SES

omniscout browser network list \
  --filter 'stripe|analytics|datadog|posthog' --session $SES

End-to-end task:
1. Use OmniScout to research "vector database benchmarks 2026" (10 results).
2. From the report, pick the 3 highest-scoring source URLs.
3. For each one: open it in OmniScout (session=audit), screenshot to
   /tmp/<i>.png, extract the markdown to /tmp/<i>.md.
4. Read all 3 markdown files and synthesize a 5-bullet summary of the
   common themes, attributing each bullet to which source(s) discussed
   it.
5. Close all OmniScout sessions when done.

export OMNISCOUT_JSON=1
omniscout daemon start

omniscout research "vector database benchmarks 2026" --results 10 > /tmp/r.json

i=1
for url in $(jq -r '.sources[:3][] | .url' /tmp/r.json); do
  omniscout browser navigate "$url" --session audit
  omniscout browser screenshot --session audit --out /tmp/$i.png
  omniscout extract "$url" > /tmp/$i.md
  i=$((i+1))
done

omniscout browser close --all

omniscout browser screenshot https://example.com --out /tmp/example.png
omniscout browser screenshot https://example.com --full-length --out /tmp/full.png
omniscout browser screenshot https://example.com --full-page --out /tmp/full.png

omniscout browser pdf https://example.com --paper a4 --out /tmp/example.pdf

# One-time login
omniscout profile create work
omniscout browser login https://app.example.com --profile work
# ...authenticate, then in another shell:
omniscout browser login --done

# Now everything else uses the same profile + session
omniscout browser navigate https://app.example.com/dashboard --profile work --session app
omniscout browser snapshot --session app
omniscout browser click '@e7' --session app
omniscout browser screenshot --session app --out /tmp/dash.png

omniscout browser navigate https://example.com --session multi
omniscout browser navigate https://news.ycombinator.com --new-tab --session multi
omniscout browser tab list --session multi
omniscout browser tab switch 0 --session multi   # back to example.com
omniscout browser tab close --session multi      # closes active tab

omniscout answer "who is the prime minister of india"
omniscout answer "what is the capital of france"
omniscout answer "how tall is mount everest" --data   # sources + retrieval path

URL=$(omniscout search "python contextvars" --limit 1 --json | jq -r '.hits[0].url')
omniscout extract "$URL" > /tmp/page.md
# Read /tmp/page.md in your editor or hand to your AI agent.

import json, subprocess

def omniscout(*args):
    r = subprocess.run(
        ["omniscout", *args, "--json"],
        capture_output=True, text=True, check=True,
    )
    return json.loads(r.stdout)

omniscout("browser", "navigate", "https://example.com")
snap = omniscout("browser", "snapshot", "--refs-only")
button = next(r for r in snap["refs"] if r["role"] == "link")
omniscout("browser", "click", button["ref"])
omniscout("browser", "screenshot", "--out", "/tmp/out.png")

#!/usr/bin/env bash
set -euo pipefail
omniscout daemon start

urls=(
  https://example.com
  https://example.org
  https://example.net
)

for url in "${urls[@]}"; do
  (
    safe=$(echo "$url" | sed 's|[^a-zA-Z0-9]|_|g')
    omniscout browser screenshot "$url" --out "/tmp/$safe.png" --session "$safe"
  ) &
done
wait

omniscout browser close --all
echo "done"

omniscout daemon status   # extension_connected should be true
omniscout browser navigate https://example.com --backend extension

export OMNISCOUT_JSON=1

omniscout search "topic"          | jq '.hits[] | {title, url}'
omniscout extract https://...      | jq '{title, word_count, published}'
omniscout extract https://... --format structured | jq '{company, pricing, twitter, docs}'
omniscout extract -q "SpaceX founder" --format structured --fields founder | jq .
omniscout research "topic"         | jq '.passages[0:3]'
omniscout browser snapshot         | jq '.refs[] | select(.role == "button")'
omniscout daemon status            | jq '.running'

mkdir -p ~/.config/omniscout
cat > ~/.config/omniscout/config.toml <<'EOF'
default_source = "hybrid"
search_limit = 20
research_results = 12
research_depth = 2
request_throttle_seconds = 0.5
embedding_model = "sentence-transformers/all-MiniLM-L6-v2"
browser = "chrome"
summary_sentences = 8
EOF

omniscout settings set browser brave
omniscout settings show

omniscout browser close --all          # close all browser sessions
omniscout daemon stop                   # stop the daemon
rm -rf ~/Library/Application\ Support/omniscout/cache/pages/*
rm ~/Library/Application\ Support/omniscout/cache.sqlite
rm -rf ~/Library/Application\ Support/omniscout/qdrant/