Sample dataset
Seven fictional accounts with cross-channel messages, source coverage and confirmed commitments.
LINKEDIN + EMAIL + CRM
Fictional sample · local analysis
THE NEXT ACTION STARTS WITH CONTEXT
Know what happened.
Decide what matters next.
One evidence-backed account view, across channels. Find unfinished commitments, missing stakeholders and the gaps that make a confident answer premature.
Accounts to review
All actions are recommendations. Priority is an editable heuristic, not likelihood to buy. The sample date is fixed so the examples remain reproducible.
FROM CONVERSATIONS TO STRATEGY
Let buyer evidence
shape the next experiment.
Inspect recurring topics, stakeholder gaps and priority weights. Topic matches are simple diagnostic patterns; confirm meaning in the original messages.
Topics in buyer messages
Counts are accounts mentioning a topic, not verified objections or causal evidence. Negation can trigger a pattern.
Priority policy
Score = industry + size + recent inbound + overdue commitments. Change the weights and inspect the reordered queue.
Buying-committee coverage
Only explicitly confirmed roles count. A title alone never establishes buying authority. Missing means absent from these records.
DESIGN FOR ECONOMICS
Spend on judgment.
Reuse the context.
A planning calculator, not a provider invoice. Enter your own verified per-call estimates. Enrichment, subscriptions, media and infrastructure are additional.
Defaults are illustrative assumptions, not quoted market prices. The estimator assumes every eligible account gets Jev and one brief; escalation replaces the routine brief. The CLI only escalates when explicitly configured.
What actually reduces cost
Before a model call
Identity checks, source freshness, opt-outs, future commitments and meeting state run in code. Research and evidence are shared across contacts in one account.
At the API boundary
The local pipeline caps request count and context size, uses an exact-context cache, reserves estimated spend before a call and makes no automatic paid retries.
A request limit is enforced; a dollar estimate is not a guaranteed billing cap. Configure provider-side spending controls separately.
BUILD STATUS, WITHOUT THE HAND-WAVING
Your sources.
One reviewable contract.
This public demo never requests keys or calls provider APIs. Live commands run locally. Raw provider data must be explicitly mapped before it becomes evidence.
What the first project does
Normalise reviewed records → reconcile evidence → prioritise accounts → expose strategy gaps → optionally classify and draft → export a context pack.
Where the second project starts
Rendering memes, HeyGen videos, ElevenLabs audio and sending follow-ups are outside this repository. The asset brief is a structured handoff, not a claim that media or outreach has happened.
Live API authentication and outputs have not been tested in this build. Connector tests use mocked HTTP responses. No enterprise adoption or measured time savings are claimed.
TAKE THE WORKFLOW WITH YOU
Inspectable code.
Reusable decisions.
Download inputs, playbooks and review packets. Run the core without an API key, then connect only the services your workflow needs.
Priority & cost policy
Your current weights, eligibility settings and model-budget estimates.
Selected account pack
Timeline, evidence IDs, gaps, recommendation and explicit execution restrictions.
Workflow contract
A versioned execution manifest for the included CLI. This file is not an n8n export.
Source & setup
API adapters, regression tests, source references, setup instructions and limitations.
Open GitHub ↗Reusable skills
Evidence reconciliation and account strategy instructions for compatible coding agents.
Browse SKILL.md files ↗