hearth.
Human-centered AI, held properly

The fire is new.
The discipline isn’t.

Hearth builds the structure that makes artificial intelligence safe enough to live with — for the oilfield, service, and fintech companies still running their most important workflows by hand.

Mission

Make AI safe enough to sit next to.

Every civilization organized itself around the same object: a contained fire. Hearth exists to build the stone that holds this one.

“AI is the new fire. Enormous energy — genuinely dangerous when loose, transformative when held in the right structure.

Dear operator,

Your team isn’t behind because they’re lazy. They’re behind because credit reviews, field tickets, invoices, and customer follow-ups were never designed to work together — and now an AI wave is arriving on top of all of it.

We don’t sell fire. We map your workflows, separate judgment from repeatable work, and build the smallest AI layer that creates real leverage — with your people signing off at every step.

— Ibrahim & Hearth

Core philosophy

Warmth is engineered. The ember, not the blaze. Structure protects.

Three ideas govern everything we design, build, and hand over.

01

Warmth is engineered

A hearth is masonry before it is comfort. We start with the workflow — the load-bearing structure — and let precision produce the warmth. Never the other way around.

02

The ember, not the blaze

A tended ember is fire at its most efficient: quiet, dense, ready. Our work is always one small, intense point of capability inside a calm structure — never glow everywhere.

03

Structure protects

The arch is humanity’s oldest trust technology: it holds weight by geometry alone. Human-in-the-loop isn’t a policy page for us. It’s drawn into the shape of the company.

Products

Parts of the hearth.

One carefully designed ecosystem. Each tool earns its place — plain, physical, and built to be handed over.

Platform

Mantle.

The operating console. Every workflow agent, artifact, and approval in one calm surface — the shelf above the fire where the day’s work sits.

See it think
Onboarding

Kindling.

The audit toolkit. We map handoffs, separate judgment from repeatable work, and find the first workflow worth automating — before anything is built.

Start with an audit
Infrastructure

Andiron.

The irons under the logs. Integrations with your ERP, CRM, email, and ledger — the quiet structure that keeps everything safely off the floor.

For technical teams
Writing

Dispatch.

Field notes from the work: what actually happened when AI met a credit committee, a field-service backlog, a month-end close. No hype. Evidence only.

Read the latest
Capabilities

Exact workflows, not vague transformation.

Each lane names a business artifact — something your team already touches every week.

FinanceCredit review packets and customer-risk summaries your analysts actually sign off onaudit first
OperationsField ticket → invoice → AR follow-up, without the Friday backlogmap gaps
VisibilityAn AI-readable business profile so buyers — and their AI tools — understand youscorecard
LeverageRepeatable analyst work separated from judgment work, on purposehuman-in-loop
How it thinks

Watch one workflow warm up.

A quarterly AR rescreen, the way it runs inside Mantle — conversation on the left, orchestration on the right.

Run the Q3 rescreen on the active portfolio. Same thresholds as last quarter.
Mantle · Credit agentUnderstood. 214 active accounts. Pulling payment behavior, exposure, and open disputes. I’ll flag anything that crosses your limits.
Mantle · Credit agentDone. 14 accounts flagged: 6 exposure breaches, 5 payment-slowdown trends, 3 missing financials. Summaries drafted for each.
Show me the three worst and draft the committee note.
Mantle · Credit agentReady for review. Nothing here is final — your sign-off is the last step, as always.
Pull portfolio dataerp · crm · email
Score against limits214 accounts
Draft account summaries14 flagged
Route to analyst● human_review
Example only — not a real client result. The agent proposes; the analyst decides.
Method

Audit first. Build small. Hand over.

Every engagement follows the same four moves — no platform lock-in, no black box.

1

Map the work

We trace tickets, invoices, credit checks, and follow-ups to where they actually break — not where the org chart says they should.

2

Separate judgment from repetition

Your analysts’ judgment is the asset. We find the repeatable work wrapped around it and automate only that.

3

Build the smallest useful layer

One workflow, one agent, one artifact. Proven in weeks, not quarters. If it doesn’t earn its place, it doesn’t ship.

4

Hand over with your people in the loop

Every output routes to a human before it becomes real. You own the system, the prompts, and the off switch.

credit_summary_q3.mdkindling · artifact
ACCT-1147exposure 118% of limitFLAG
ACCT-0892days-late trending +9FLAG
ACCT-1204financials 14mo staleFLAG
ACCT-0776within all thresholdsok
Drafted by agent · reviewed by analyst · Example only — not a real client result.
Technology

Boring where it should be. Careful where it counts.

The stack is chosen for longevity and auditability — the same way you’d choose materials for a building.

01 / models

Frontier models, held loosely

We route tasks to the model that fits — and keep every prompt, retrieval source, and output logged so no vendor becomes a load-bearing wall.

02 / data

Your data stays yours

Private indexes on your infrastructure or ours under agreement. No training on client data. Retention you can audit. Plain-language answers on request.

03 / control

An off switch that works

Every agent has a human gate, a rollback, and a kill path. If the system can’t explain a decision, it doesn’t make one — it asks.

Case studies

The shape of the work.

Three representative engagements. Specifics shared under NDA; patterns told plainly.

Oilfield services

From field ticket to invoice in one day

A 40-truck service company ran tickets on paper and closed invoices weekly. We mapped the handoff, put one agent on ticket-to-draft-invoice, and kept the controller as the final gate.

Pattern: operations handoff · Illustrative, not a client result
Fintech / credit

Analyst leverage on the quarterly rescreen

A specialty lender’s analysts spent the last week of each quarter assembling review packets. One workflow agent now drafts every packet; analysts review, adjust, and sign.

Pattern: finance artifact · Illustrative, not a client result
Industrial services

A website buyers’ AI can read

A regional distributor was invisible to the AI tools its buyers had started using. We rebuilt their service pages and entity profile so both humans and machines understand the business.

Pattern: AI visibility · Illustrative, not a client result
Developers

Built in the open, handed over whole.

Technical teams get the same thing our clients do: systems they can read, run, and own.

The Hearth agent SDK

from hearth import agent, gate

credit = agent(
    task="ar_rescreen",
    sources=["erp", "crm", "email"],
    review=gate("analyst"),  # always
)

Agents are declared with their data sources and their human gate in one place. No gate, no run. That’s not a guideline — it’s enforced.

Request the technical overview

Self-hosted by default

$ hearth audit ./workflows/ar.yaml
  mapped: 14 steps · 3 handoffs
  judgment: 2 steps (analyst-owned)
  automatable: 6 steps
  est. leverage: 11.5 hrs / week

Kindling runs against your own exports — no credentials leave your network during an audit. You get the map before we talk about building anything.

Book a working session
Careers

Sit closer to the fire.

We hire people who have done the work by hand and respect it — operators, analysts, and engineers who want their craft to outlast the demo. If that’s you, the door is open.

Introduce yourself
Contact

Start with one workflow that’s leaking time.

The AI Modernization Audit takes two weeks and ends with a map, a shortlist, and a build plan you can execute with us or without us. No obligation past the map.