Keep This Moving: Momentum on HUMΛN

You finish a run in Companion on Friday. It works. Monday you type the same prompt again—because the session died and nothing persisted the goal.

That is not a you problem. It is a lifecycle problem. Momentum is HUMΛN’s answer: persistence with policy, not a calendar reminder glued to a chat transcript.

Scroll-stopper: If recurring work only lives in your memory of last week’s prompt, you do not have a system—you have a habit waiting to fail.

The problem

Most AI tools optimize for the session. You ask, it answers, the context evaporates. Recurring work becomes copy-paste prompts, calendar reminders, or brittle third-party chains that nobody trusts with delegation, risk gates, or provenance.

HUMΛN already routes one-shot work through Intent → capabilities → HumanOS execution. Momentum adds the missing layer so successful work can keep moving under the same governance stack—so that Monday does not require re-explaining Friday’s intent.

How Momentum works

A Momentum object stores:

Goal — what to keep doing, in plain language
Trigger — cron, condition, or manual
Capabilities — which connectors execute the work
Delivery — where results land (Companion, Slack, email)
Policies — approval gates, cost caps, notification urgency

On each trigger, the executor compiles your goal, runs it through the same governance stack as any human.call, and records a provenance-backed run. Condition-based Momentums compare output fingerprints so you only hear about meaningful change.

Magic by default

Most developers never touch cron syntax:

const obj = await client.momentum
  .build('Morning sales brief from HubSpot')
  .every('weekdays at 8am')
  .using(['hubspot', 'slack'])
  .delivering_to('slack')
  .create();

every('weekdays at 8am') becomes a cron expression. Templates (morning-brief, pipeline-scout, watch-this-for-me, …) pre-fill triggers and capability families. The CLI mirrors the SDK: human momentum create, run --follow, logs --follow.

Control mode is always there — explicit triggers, policies, and CreateMomentumOptions for the ten percent who need it. Magic by default, control when needed.

Governance and human-in-the-loop

Momentum is not unattended “set and forget” automation wearing a trust badge. Approval policies, awaiting_human run states, and pause/resume keep humans sovereign (Principle #1) and in the loop (Principle #5). Low-confidence routing defaults to human review, not silent failure. Every run still produces a provenance trail.

Use cases from the template library

Template	What it does
Morning Brief	Calendar + inbox + priorities every weekday morning
Pipeline Scout	CRM and social lead signals in a broad daily window
Watch This For Me	Condition monitor — notify only on meaningful change
Reclaim My Week	Friday focus-time and calendar optimization
Social Signal Scout	Keyword and account monitoring for lead signals
Local Event Scout	Location-based event and ticket alerts

Try it

Getting Started guide — SDK + CLI walkthrough
Quickstart runbook — fifteen lines to first run
MCP tools guide — call human.momentum.create from Claude or Cursor

After a successful Intent run, Companion will offer Keep this moving → — one tap to pre-fill a Momentum object from what already worked.

That is the point: stop re-explaining Monday’s work every Monday.

Go deeper

Product: HumanOS · Companion
Docs: Introduction · Guides
Community: Agents before/after trust layer · API designed for builders · Companion from Claude Code