Deepline Quickstart

Run a high-confidence demo recipe to show the user what Deepline can do. Pick the most relevant recipe below, or default to Recipe 1 if no context is given.

Always prefer the hardcoded recipes below. /gtm-meta-skill is always available as a fallback but should only be used if: (a) a recipe command fails and all fallbacks are exhausted, or (b) the user's ask doesn't match any recipe here. Never invoke it preemptively.

Execution flow

Follow this pattern for every recipe:

Tell the user what you're about to do — explain the goal and which data source(s) you'll use, before running anything.
Register a plan with deepline session plan --steps '[...]' matching the recipe steps.
For each step: mark it running, send a live status message describing what's happening, run the command, then mark it completed (or error on failure).
Register output with deepline session output --csv <path> --label "..." after any CSV is produced.
Tell the user the results — summarize what came back, where it came from, and what they can do next.

Session commands reference

deepline session plan --steps '["Step 1", "Step 2"]'
deepline session plan --update <i> --status running|completed|error|skipped
deepline session status --message "What's happening right now..."
deepline session output --csv <path> --label "Label for the table"

Recipe 1 — Find CTOs at NY startups

Goal: Find 5 CTOs at startups in New York with verified emails and LinkedIn profiles. Data sources: Dropleads (people search) + waterfall email enrichment via person_linkedin_only_to_email_waterfall.

Steps:

Search Dropleads for CTOs in New York
Waterfall enrich emails
Display results

Step 1 — Search

deepline tools execute dropleads_search_people --payload '{
  "filters": {
    "jobTitles": ["CTO"],
    "personalStates": {"include": ["New York"]},
    "employeeRanges": ["1-10", "11-50", "51-200"]
  },
  "pagination": {"page": 1, "limit": 5}
}'

Note the output CSV path from the result.

Step 2 — Waterfall enrich emails

First, prep the name and LinkedIn columns the play expects:

deepline enrich --input <csv_from_step_1> --in-place \
  --with '{"alias":"first_name","tool":"run_javascript","payload":{"code":"return (row[\"fullName\"]||\"\").trim().split(\" \")[0]||null;"}}' \
  --with '{"alias":"last_name","tool":"run_javascript","payload":{"code":"const parts=(row[\"fullName\"]||\"\").trim().split(\" \"); return parts.slice(1).join(\" \")||null;"}}' \
  --with '{"alias":"linkedin_url","tool":"run_javascript","payload":{"code":"return row[\"linkedinUrl\"]||null;"}}'

Then run the waterfall play:

deepline enrich --input <csv_from_step_1> --in-place \
  --with '{"alias":"email","tool":"person_linkedin_only_to_email_waterfall","payload":{"linkedin_url":"{{linkedin_url}}","first_name":"{{first_name}}","last_name":"{{last_name}}"}}'

Step 3 — Display results

Show a summary table: name, company, email, LinkedIn URL. Tell the user emails were filled via waterfall enrichment across Dropleads, Deepline Native, Crustdata, and PDL. Mention they can go deeper — phone, firmographics, job change signals — with /gtm-meta-skill.

Fallback (if Step 1 errors)

Tell the user, then try Apollo:

deepline tools execute apollo_search_people_with_match --payload '{
  "person_titles": ["CTO", "Chief Technology Officer"],
  "person_seniorities": ["c_suite"],
  "person_locations": ["New York, New York, United States"],
  "organization_num_employees_ranges": ["1,200"],
  "include_similar_titles": true,
  "per_page": 5,
  "page": 1
}'

Last resort

If all commands fail, tell the user, then invoke /gtm-meta-skill:

Find 5 CTOs at startups in New York with their emails and LinkedIn profiles.

deepline-quickstart