saber-cli

name

saber-cli

description

GTM intelligence via the Saber CLI. Use when the user wants to define their ICP and buying signals, build target account or contact lists, run company or contact signal research, manage signal subscriptions, or prioritize outreach. Also use when the user mentions prospecting, lead scoring, intent data, buying intent, firmographic filters, or anything related to sales intelligence. Triggers include "find companies", "who should I target", "track hiring signals", "build a list", "run signals", "check credits", or any GTM workflow involving the saber command.

Saber CLI

The Saber CLI runs company and contact signal research, builds target lists, and manages signal subscriptions from the terminal. It talks to the Saber API, authenticated with an API key stored at ~/.saber/credentials.json.

Core Workflow

Every Saber engagement follows three phases:

1. Discover -- define ICP and buying signals
2. Build -- create target account and contact lists
3. Activate -- run signals against those lists

# Phase 1: Understand the org
saber org get

# Phase 2: Build a company list
saber list company create --name "Mid-market SaaS" --industry "software development" --size "51-200" --country US

# Phase 3: Run a signal across the list
saber subscription create --list <listId> --name "Hiring in sales" \
  --question "Is this company actively hiring in sales or revenue roles?" \
  --answer-type boolean --frequency monthly --run-once

Phase 1 -- Signal Discovery

Help the user define what buying intent looks like for their Ideal Customer Profile before activating any signal tracking.

Step 1 -- Load organisation context

Run saber org get first. If the profile has name, website, and description fields, use them directly. If the profile is empty, ask the user to fill the gaps, then persist:

saber org update --name "Acme Corp" --general "B2B sales intelligence platform" \
  --products "Signal tracking, list building" \
  --use-cases "Outbound prospecting, ABM campaigns" \
  --value-prop "Find buying intent before your competitors do"

Step 2 -- Gather ICP details

Once org context is established, ask these in order (conversationally, not all at once):

1. Who do you sell to? -- industry, company size, geography, buyer title
2. Who is the buyer? -- title, seniority, department (e.g. "VP Sales", "Head of RevOps")
3. What makes a company ready to buy? -- triggers (new hire, funding, tech migration)

Step 3 -- Generate signal hypotheses

Propose 3-5 signals based on the ICP. For each:

• Signal question -- natural language question Saber will answer
• Signal type -- company or contact
• Why it matters -- brief rationale linking the signal to purchase readiness

Example signals:

Signal: Is this company actively hiring in sales or revenue roles?
Type: company
Rationale: Hiring in sales indicates growth mode and budget for new tools.

Signal: Has this company recently raised a funding round?
Type: company
Rationale: Post-funding companies accelerate spend on GTM tooling.

Signal: Is the Head of People posting about frontline retention challenges?
Type: contact
Rationale: Public pain signals from the buyer persona indicate active problem awareness.

Step 4 -- Confirm and hand off

Present proposed signals, adjust based on feedback, then keep the agreed ICP and signal definitions in conversation context. Next steps: build lists (Phase 2), then activate signals (Phase 3).

Phase 2 -- Build Target Lists

Company lists

Ask how the user wants to supply accounts:

Option A -- Let Saber find companies (recommended)

saber list company create --name "US Mid-market SaaS" \
  --industry "software development" --size "51-200" --country US

Filter formats:

• --industry values are lowercase: restaurants, hospitality, food & beverages
• --size uses K notation: 1-10, 11-50, 51-200, 201-500, 501-1K, 1K-5K, 5K-10K, 10K+
• --country uses ISO 3166-1 alpha-2: US, GB, DE, NL
• --technology uses slugs: stripe, hubspot, salesforce

Technology filters consume credits per matched company. Always run count-preview first so the user can see the cost before committing:

saber list company count-preview --technology "salesforce" --size "51-200" --country US
# Shows: X companies matched, Y credits required
# Ask user to confirm before running create

Option B -- Import from HubSpot

saber list company import --name "From HubSpot" \
  --property industry --operator EQ --value "Technology"

Preview without creating:

saber company search --industry "software development" --size "51-200" --country US

Contact lists

Contacts are sourced from company LinkedIn URLs. If a company list exists, get URLs from it first:

saber list company companies <listId> --limit 10

Then create the contact list:

saber list contact create --name "VP Sales at target accounts" \
  --company-linkedin https://linkedin.com/company/acme \
  --company-linkedin https://linkedin.com/company/globex \
  --title "VP Sales" --title "Head of Sales"

Both --company-linkedin and --title are repeatable. --company-linkedin is required.

Preview without creating:

saber contact search --company-linkedin https://linkedin.com/company/acme --title "VP Sales"

Filter by department and seniority using the LinkedIn Sales Navigator taxonomy (both repeatable), and page through large result sets with --limit/--offset. Values must match the LSN taxonomy exactly — e.g. seniority is Vice President, not VP; unknown values are rejected by the API with a validation error:

saber contact search --company-linkedin https://linkedin.com/company/acme \
  --department Sales --seniority CXO --seniority "Vice President" \
  --limit 50 --offset 0

Look up a single contact's email (one-shot, given just full name + company domain):

saber contact find-email --full-name "Joey van Ommen" --domain saber.app

Reach for find-email when you already have a target name in mind — e.g. enriching a list with verified emails before drafting outreach, or resolving a contact you saw on LinkedIn. It's distinct from contact search (which previews many contacts by filter) and from list contact create (which snapshots a Sales Navigator list). Catch-all domains return the modal real-world pattern with accept_all: true — treat as lower confidence.

Market Signals

Market signals monitor external data sources continuously and deliver matching signals to a webhook. Unlike company/contact signals which answer research questions, market signals track real-time events: job postings, LinkedIn posts, fundraising, investments, and IPOs.

Subscription types

Type	Description
JOB_POSTS	Monitor job postings matching title, location, company filters
LINKEDIN_POST	Monitor LinkedIn posts matching keyword filters (requires Sales Navigator)
FUND_RAISED	Monitor fund closings and fundraising events
RECENT_INVESTMENT	Monitor investment and funding rounds
IPO	Monitor IPO and public listing announcements

Create a market signal subscription

# Monitor DevOps job postings in the US
saber market-signal create --type JOB_POSTS --name "DevOps hiring" \
  --filters '{"titleKeywords":["DevOps Engineer","SRE"],"countries":["US"]}' \
  --webhook https://example.com/hook

# Use AI to generate filters from a prompt
saber market-signal create --type JOB_POSTS --name "AI startups" \
  --prompt "Find companies hiring ML engineers at Series A startups in Europe" \
  --webhook https://example.com/hook

# Monitor LinkedIn posts about AI
saber market-signal create --type LINKEDIN_POST --name "AI posts" \
  --filters '{"keywordsAll":["artificial intelligence"],"keywordsAny":["GPT","LLM"]}' \
  --webhook https://example.com/hook

# Monitor fundraising events weekly
saber market-signal create --type FUND_RAISED --name "PE funds" \
  --filters '{"keywords":["private equity"],"maxLookbackDays":30}' \
  --webhook https://example.com/hook --interval weekly

Manage market signal subscriptions

saber market-signal list                          # List all subscriptions
saber market-signal get <subscriptionId>          # Get subscription details
saber market-signal update <id> --name "New name" # Update subscription
saber market-signal pause <id>                    # Pause polling
saber market-signal resume <id>                   # Resume polling
saber market-signal trigger <id>                  # Trigger immediate run
saber market-signal signals <id>                  # List matched signals
saber market-signal delete <id>                   # Delete subscription

The alias saber ms works as shorthand for saber market-signal.

Phase 3 -- Activate Signal Tracking

Before running any signals

Before any saber signal or saber subscription command that consumes credits:

1. Run saber credits and show the current balance. Balances can be whole or fractional credits.
2. State how many signals will run (companies x questions, or contacts x questions)
3. Ask the user to confirm before proceeding

Signal templates

Create reusable templates for frequently used signal questions. Templates store the question, answer type, weight, and qualification criteria so they can be referenced by ID instead of repeating the full definition.

# Create a template
saber template create --name "CRM Detection" --question "Which CRM are they using?" \
  --answer-type list --weight important

# List templates
saber template list

# Get template details
saber template get <templateId>

# Update a template (creates a new version)
saber template update <templateId> --name "CRM Detection v2" --weight nice_to_have

# Delete a template (soft-delete)
saber template delete <templateId>

Extract templates from historical ad-hoc signals

If an org has been running ad-hoc signals (saber signal --question ...) before templates existed, those executions can be retroactively grouped into reusable templates so they're scoreable. The flow is two steps:

# 1. Propose: cluster historical ad-hoc signals into template proposals.
#    The LLM groups equivalent questions and matches against existing templates
#    rather than proposing duplicates.
saber template extract propose --type company --json > plan.json

# 2. Review plan.json — drop unwanted clusters, tweak names/questions.
#    Each cluster is either kind="new" (create a fresh template) or
#    kind="existing" (attach executions to a template that already covers
#    the same intent).

# 3. Apply: creates templates and back-fills the executions in one transaction.
saber template extract apply --from-file plan.json

Or pipe straight through if you trust the proposal as-is. --yes skips the credit-confirmation prompt that propose would otherwise block on:

saber template extract propose --type contact --yes --json | \
  saber template extract apply --from-file -

Re-running the same plan returns 409 — drop already-attached executionIds before retrying. Only score-compatible answer types (boolean, number, percentage, currency, list) are clustered; open-text answers are skipped.

Prebuilt signals (use when one answers the question exactly)

Five prebuilt signals run by key with no question design needed: funding, mna, tech, open-jobs, firmographics. Each answers one commodity fact — same credit confirmation, same output, same --no-wait/--webhook async options.

Use one only when the question IS that fact, with no threshold, window, filter or judgement attached. "Has this company raised recently?" is funding; "raised a Series B+ in the last 6 months and has no VP Sales yet" is a custom --question — a prebuilt key cannot express it, and reshaping the question to fit one throws away the qualification criteria that make it worth asking.

saber signal funding --domain acme.com
saber signal tech --domain acme.com --category crm        # or --technology "salesforce"
saber signal mna --domain acme.com --no-wait

tech takes exactly one of --category erp|crm or --technology "<name>"; an unknown technology returns a 422 with did-you-mean suggestions.

Company signals -- spot check

Run a question against a single domain:

saber signal --domain acme.com --question "Are they hiring engineers?" --answer-type boolean

Use a template instead of inline question:

saber signal --domain acme.com --template <templateId>

Use lenient verification mode for broader answers:

saber signal --domain acme.com --question "What is their tech stack?" --verification-mode lenient

Fire multiple in parallel with --no-wait:

saber signal --domain acme.com --question "Are they hiring?" --no-wait
saber signal --domain globex.com --question "Are they hiring?" --no-wait
# Collect signal IDs, then retrieve:
saber signal get <signalId>

Company signals -- batch

Run multiple questions across multiple domains in one call. Creates a Cartesian product (each question x each domain).

# 2 questions x 3 domains = 6 signals
saber signal batch \
  --domain acme.com --domain globex.com --domain initech.com \
  --question "Are they hiring engineers?" \
  --question "What CRM do they use?" \
  --answer-type boolean

# Auto-generate summaries when all signals complete
saber signal batch --domain acme.com --domain globex.com \
  --question "Revenue?" --question "Headcount?" --generate-summary

# Async mode for large batches (up to 20,000 signals)
saber signal batch --domain acme.com --question "..." --async

Company signals -- listing and filtering

Browse and filter existing signal results:

saber signal list                                          # all signals
saber signal list --domain acme.com --status completed     # filter by domain and status
saber signal list --from-date 2024-01-01T00:00:00Z         # filter by date
saber signal list --subscription-id <id>                   # filter by subscription
saber signal list --limit 10 --offset 20                   # pagination

Signal summaries

Generate AI-powered summaries that consolidate insights from all completed signals for a domain into structured data points with qualifications and sources.

# Generate a new summary
saber summary generate --domain acme.com

# List historical summaries for a domain
saber summary list --domain acme.com

Company signals -- full list (subscriptions)

For running a signal across all companies in a list, use subscriptions.

One-off run (recommended for getting started):

saber subscription create --list <listId> \
  --name "Hiring signal" \
  --question "Is this company actively hiring in sales or revenue roles?" \
  --answer-type boolean \
  --frequency monthly --run-once

--run-once triggers immediately, then stops the schedule. One subscription per signal question.

Recurring schedule:

saber subscription create --list <listId> \
  --name "Hiring signal" \
  --question "Is this company actively hiring in sales or revenue roles?" \
  --answer-type boolean --frequency weekly

Structured data (json_schema):

saber subscription create --list <listId> \
  --name "Company financials" \
  --question "Extract annual revenue, employee count, and funding stage" \
  --answer-type json_schema \
  --output-schema '{
    "type": "object",
    "properties": {
      "revenue": {"type": "string"},
      "employees": {"type": "number"},
      "funding_stage": {"type": "string"}
    }
  }' \
  --frequency monthly

Or use a schema file:

saber subscription create --list <listId> \
  --name "Company financials" \
  --question "Extract annual revenue, employee count, and funding stage" \
  --answer-type json_schema \
  --output-schema @schema.json \
  --frequency monthly

Manage subscriptions:

saber subscription list
saber subscription get <subscriptionId>
saber subscription trigger <subscriptionId>    # run immediately
saber subscription start <subscriptionId>      # activate schedule
saber subscription stop <subscriptionId>       # pause schedule

Contact signals

Run a question against a LinkedIn profile:

saber signal --profile https://linkedin.com/in/janedoe \
  --question "Is this person posting about employee retention challenges?"

Use a template:

saber signal --profile https://linkedin.com/in/janedoe --template <templateId>

Fire multiple with --no-wait:

saber signal --profile <linkedin-url> --question "<question>" --no-wait
saber signal get <signalId>

Sample run (recommended before full activation)

Before running signals across a full list, test a sample of 5-10 items first so the user can validate quality without spending credits on everything.

For company signals:

saber list company companies <listId> --limit 10
# For each domain, fire signals with --no-wait
saber signal --domain <domain> --question "<question>" --answer-type boolean --no-wait
# Collect IDs, then retrieve
saber signal get <signalId>

Or use batch for a quicker multi-domain sample:

saber signal batch --domain acme.com --domain globex.com --domain initech.com \
  --question "Are they hiring?" --answer-type boolean

Present results and ask:

• Do the signals look accurate?
• Any false positives or unexpected answers?
• Proceed with the full list?

For contact signals, pick 3-5 contacts the user already knows about (so they can judge accuracy), run signals, review, then confirm.

Scoring

Scoring turns signal results into structured fit and urgency scores (0-100) for companies and contacts. It builds on top of signals -- any data the signal layer can ingest becomes scoreable.

Mental model

1. Profile -- a named, org-scoped configuration bound to a single object type (company or contact). Holds rules. One object can be assigned to multiple profiles of the matching type.
2. Rule -- within a profile, maps one signal template to point values for one dimension (fit or urgency). The shape of point values depends on the template's answer type.
3. Assignment -- links a profile to a specific company or contact. Triggers an immediate recompute.
4. Score result -- the computed score for one (profile, object, dimension) triple, with a per-rule contribution breakdown so it's always explainable.

Score is (earned / maxPossible) * 100. All rules count toward the denominator regardless of whether a signal answer exists, so low-coverage scores stay conservatively low.

Manage profiles

saber scoring profile create --type company --name "EMEA Enterprise" \
  --description "1000+ headcount in EMEA"
saber scoring profile list
saber scoring profile get <profileId>
saber scoring profile update <profileId> --name "Enterprise ICP"
saber scoring profile delete <profileId>          # cascades to rules, assignments, scores

Profile --type is immutable after creation.

Manage rules

A rule's point-value shape depends on the answer type of the referenced signal template:

# Boolean signal
saber scoring rule upsert <profileId> --signal-template <id> --dimension fit \
  --true 20 --false -5

# Number / percentage / currency signal -- ranges (max is exclusive)
saber scoring rule upsert <profileId> --signal-template <id> --dimension fit \
  --range "0:500:5" --range "500:5000:15" --range "5000:100000:25"

# List signal
saber scoring rule upsert <profileId> --signal-template <id> --dimension fit \
  --choice "Salesforce:10" --choice "HubSpot:8" --choice "None:-10"

# Or pass raw JSON / a JSON file matching the ScoringPointValues schema
saber scoring rule upsert <profileId> --signal-template <id> --dimension urgency \
  --points '{"ranges":[{"min":0,"max":500,"points":5}]}'
saber scoring rule upsert <profileId> --signal-template <id> --dimension urgency \
  --points-file rules.json

saber scoring rule list <profileId>
saber scoring rule delete <profileId> <ruleId>     # also recomputes assigned objects

Upserting or deleting a rule recomputes every object assigned to the profile, so scores never go silently stale.

Assign profiles to objects

Object IDs are domains for companies and LinkedIn profile URLs for contacts.

# One object
saber scoring assignment create --profile <id> --type company --object acme.com

# Many at once -- duplicates are skipped, only new assignments are returned
saber scoring assignment bulk --profile <id> --type company \
  --object acme.com --object stripe.com --object framer.com

# Profiles assigned to a single object
saber scoring assignment list --type company --object acme.com

saber scoring assignment delete <assignmentId>     # also clears its scores

Read and recompute scores

# Compact table across all matching profiles + dimensions
saber scoring scores --type company --object acme.com

# Multiple objects in one call
saber scoring scores --type company --object acme.com --object stripe.com

# Detailed view with per-rule contribution breakdown
saber scoring scores --type company --object acme.com --detailed

# Trigger an async recompute (idempotent -- duplicate triggers attach to the
# running workflow rather than starting a new one). Read results with `scores`.
saber scoring compute --type company --object acme.com --object stripe.com

Recompute is also triggered automatically on profile assignment and on rule upsert/delete -- direct compute calls are mainly for refreshing against the latest signal data.

Signal Flags

Flag	Default	Description
--domain / -d	--	Company domain (company signals)
--profile / -p	--	LinkedIn profile URL (contact signals)
--question / -q	required*	Research question (max 500 chars)
--template	--	Signal template ID (alternative to --question)
--answer-type / -a	open_text	open_text, boolean, number, list, percentage, currency, url, json_schema
--output-schema	--	JSON Schema string or @file path (required when answer-type is json_schema)
--verification-mode	strict	strict or lenient
--force-refresh	false	Bypass 12h result cache
--no-wait	false	Return signal ID immediately without polling
--webhook	--	POST result to this URL when complete
--max-wait	120	Max seconds to wait before timing out

*Required unless --template is provided.

Common Patterns

End-to-end ICP activation

# 1. Set up org context
saber org update --name "Acme" --general "We sell HR software to mid-market"

# 2. Build target list
saber list company create --name "Mid-market US" \
  --industry "software development" --size "51-200" --country US

# 3. Sample test (10 companies)
saber list company companies <listId> --limit 10
saber signal --domain <domain> --question "Are they hiring in HR?" --no-wait

# 4. Full activation
saber subscription create --list <listId> --name "HR hiring" \
  --question "Is this company actively hiring in HR roles?" \
  --answer-type boolean --frequency monthly --run-once

# 5. Review results
saber subscription get <subscriptionId>

Multi-signal activation

Create one subscription per signal question against the same list:

saber subscription create --list <listId> --name "Hiring signal" \
  --question "Is this company actively hiring?" --answer-type boolean --frequency monthly --run-once
saber subscription create --list <listId> --name "Funding signal" \
  --question "Has this company raised funding recently?" --answer-type boolean --frequency monthly --run-once
saber subscription create --list <listId> --name "Tech migration" \
  --question "Is this company migrating their tech stack?" --answer-type boolean --frequency monthly --run-once

Or use batch for a quick multi-question run across specific domains:

saber signal batch \
  --domain acme.com --domain globex.com \
  --question "Is this company actively hiring?" \
  --question "Has this company raised funding recently?" \
  --question "Is this company migrating their tech stack?" \
  --answer-type boolean --generate-summary

Template-based workflow

Create templates once, reuse across signals and subscriptions:

# Create templates
saber template create --name "Hiring" --question "Is this company actively hiring?" --answer-type boolean
saber template create --name "Funding" --question "Has this company raised funding recently?" --answer-type boolean

# Run signals using templates
saber signal --domain acme.com --template <templateId>

# Use templates in subscriptions
saber subscription create --list <listId> --template <templateId> --frequency weekly

Check account status

saber credits                    # remaining credit balance, including fractional credits
saber connectors                 # configured integrations and status
saber auth status                # current auth state
saber update                     # check for newer version and upgrade

Update checks

The CLI checks for updates in the background (once per 24h, non-blocking). A one-line notice is printed to stderr when an update is available. The check is skipped when --json, --quiet, or SABER_NO_UPDATE_CHECK=1 is set, when stderr is not a TTY, or on dev builds.

Global Flags

Available on every command:

Flag	Description
--json	Output raw API JSON to stdout
--quiet / -Q	Suppress all non-error output
--verbose / -v	Log HTTP method, URL, status, rate-limit headers to stderr
--api-url	Override base API URL (default: https://api.saber.app)

Deep-Dive Documentation

Reference	When to Use
references/cli-reference.md	Full command reference with all flags and options
references/signal-examples.md	Signal question examples by industry and use case