AI product company

We build the AI that runs your industry.

Herantes designs, builds and operates end-to-end AI technology for business: research through runtime, data through deployment. Then we ship it as products: Sanaris for healthcare, LegalFront for legal, Lightng for business.

Book a demoWatch the film
3
Products shipped and live
4–6 wks
Concept to production
500+
Firms on LegalFront
SOC 2
HIPAA aligned, GDPR ready
Frontier & open modelsRetrieval & vector searchAgent runtimeTool orchestrationVoice & telephonyEvaluation & guardrailsFine-tuningStreaming pipelinesMulti-cloud deploymentObservabilityHuman-in-the-loop reviewCompliance & audit trails

The products

Three industries. Three products. One platform.

Every one of them runs on the same Herantes stack. We built the platform first, then proved it three times over in the three industries where the work is hardest and the stakes are highest.

01 / Healthcare

Sanaris

The record that keeps up with the round.

The clinician-first EHR. Sanaris listens to the visit and drafts the structured, billable note: diagnosis, plan and orders already in place. One adaptive chart replaces tab-hunting across modules, and FHIR and HL7 interoperability ship on day one.

  • Ambient notes, written for you
  • One screen, the whole patient
  • Orders in three taps
  • FHIR & HL7 from day one
ehr.herantes.com, visit Sanaris
~0 hrs
Documentation returned daily
0 screen
Whole patient, no modules
Day one
FHIR & HL7 interoperability
02 / Legal

LegalFront

Legal work, reimagined.

The AI workspace built for modern law firms. Contract intelligence that extracts fifty-plus key terms in under thirty seconds, citation-verified research grounded in authoritative sources, automatic deadline extraction, and due diligence at M&A scale.

  • Contract intelligence
  • Citation-verified research
  • Deadline management
  • Due diligence at M&A scale
legalfront.herantes.com, visit LegalFront
0%
Time saved on contract review
0+
Law firms on the platform
<0s
Average contract analysis
03 / Business

Lightng

The CRM that works the pipeline with you.

Boards your team will actually keep up to date, email that goes out from your own address personalised per contact, and the reporting to show what it earned. Shelly, the built-in assistant, reads what you ask for and does it.

  • Boards that fit your data
  • Send from your own mailbox
  • Shelly, the AI assistant
  • Revenue reporting per send
lightng.site, visit Lightng
0 boards
Ready on day one
0 days
Free trial, no card
0 tool
For the whole sales cycle

Ninety seconds

What we actually build, without the deck.

The platform

End to end means all five layers.

Most vendors hand you one layer and a diagram of the rest. Herantes builds and operates all five, which is the only reason we can put three products of our own into production on top of them.

01 / Research & models

The right model for the job, proven before it ships

We benchmark frontier and open-weight models against your actual task, not a public leaderboard. Where a smaller fine-tuned model wins on cost and latency, we say so, and we build the evaluation harness that keeps proving it after launch.

  • Task-specific evaluation sets built from your data
  • Frontier and open-weight models benchmarked side by side
  • Fine-tuning and distillation where it pays for itself
  • Regression suites that run on every model change
eval / contract-clause-extraction · n=1,240
ModelAccCostP95
frontier-lg
94%100100
frontier-sm
91%2234
open-8b-ft← SHIPPED
93%618
open-8b-base
71%618

Fine-tuned open model matches frontier accuracy at 6% of cost. Cost and P95 indexed to frontier-lg = 100.

02 / Agent runtime

Agents that decide, act, and hand back when they should

A durable runtime with real tool calling, retries, and state that survives a restart. Every run is traced end to end, every decision is inspectable, and escalation to a human is a first-class path rather than an error case.

  • Durable execution with replay and resume
  • Typed tool calling against your own systems
  • Human-in-the-loop escalation as a designed path
  • Full trace of every decision, replayable
trace / run_8f2c41 · agent=receptionist
  1. 0.00sTRIGGERinbound call · +1 415 ···
  2. 0.31sRETRIEVEaccount, 3 open tickets
  3. 0.88sTOOLcrm.lookup(contact_id)
  4. 1.42sTOOLcalendar.find_slot(2d)
  5. 2.10sDECIDEconfidence 0.94 · proceed
  6. 2.64sCOMMITbooking written · SMS sent
Escalationnot required · threshold 0.80

03 / Data & retrieval

Grounded in your data, with a citation for every claim

Ingestion, chunking, embedding and hybrid retrieval tuned per corpus, because contract clauses, clinical notes and CRM records do not chunk the same way. Answers carry citations back to source, which is what makes them usable in regulated work.

  • Hybrid semantic and keyword retrieval
  • Per-corpus chunking and embedding strategy
  • Citations resolved to the source document
  • Permission-aware retrieval, per user and per role
retrieval / corpus=clinical-notes · k=4
0.91discharge_summary_2024-11.pdf§4
0.88medication_history.fhirobs 12
0.84consult_note_cardiology.pdfp2

Every answer resolves to a source and a location.

04 / Interfaces

Wherever the work already happens

The best AI system is worthless if people have to leave their workflow to reach it. We ship the surface too: web, mobile, voice, and embedded inside the tools your team already has open all day.

  • Web and mobile applications, built in-house
  • Sub-second voice agents on real telephony
  • Embedded in Slack, Teams, and your CRM
  • APIs and webhooks for everything else
surfaces / one system, four front doors

Web

React app, your brand

Mobile

iOS and Android

Voice

Sub-second on real PSTN

Embedded

Slack, Teams, CRM

Live voice session340 ms to first word

05 / Deploy & operate

We stay on after launch

Deployed to your cloud or ours, with the monitoring, cost controls and on-call that keep it running. Most engagements go from concept to production in four to six weeks, and we operate what we build.

  • Your VPC, our cloud, or on-premise
  • Latency, cost and quality monitored per run
  • Alerting and on-call rotation included
  • Four to six weeks, concept to production
operate / production · last 30 days

99.97%

Uptime

412 ms

P95 latency

$0.011

Cost per run

Cost per run, trending−42%
us-easteu-westap-southyour VPC

In production

The numbers come from our own shipped products.

LegalFront

0%

Time saved on contract review

Sanaris

~0 hrs

Documentation returned per clinician, per day

LegalFront

0+

Law firms in production

0.0%
Citation accuracy, verified
<0s
Full contract analysis
0 ms
P95 agent response
0.00%
Production uptime
0–6 wks
Concept to production
0
Products of our own, live

We do not sell you a pilot and leave. We build the system, put it in production, and stay on the pager.

How every Herantes engagement is structured

How we work

Four to six weeks, and then we stay.

  1. 01Week 1

    Audit

    We sit with the people doing the work and map where the time actually goes. You leave with a scored one-page readout and the two workflows with the fastest payback.

    Scorecard + fixed quote

  2. 02Weeks 2–3

    Prototype

    A working system on your data, in your environment, evaluated against your task. Not a slide, not a sandbox. Something your team can break.

    Working prototype + eval set

  3. 03Weeks 4–6

    Production

    Hardening, integrations, permissions, escalation paths and observability. Deployed to your cloud or ours, with the runbook written.

    Live system + runbook

  4. 04Ongoing

    Operate

    We monitor latency, cost and quality per run, retrain when the data shifts, and stay on-call. The system gets better after launch, not worse.

    On-call + monthly review

Security & compliance

Built to pass your security review.

We build for hospitals and law firms. The controls this kind of work belongs under are not an enterprise upsell here. They are the default, designed in from the first commit.

SOC 2 Type IIHIPAA / BAAGDPRCCPAISO 27001Air-gapped option
Request the security pack99.95% uptime SLA · 24/7 support
Your data never trains a model
No customer data enters any training or fine-tuning run without a separate, written agreement. Processing happens in isolated, ephemeral environments.
Encrypted, isolated, deletable
AES-256 at rest and in transit, per-tenant isolation, and instant deletion on request. Sanaris runs under a signed BAA; LegalFront holds SOC 2 Type II.
Deployed where you need it
Your VPC, our cloud, or on-premise. Regional data residency in US, EU and APAC, and an air-gapped option for the work that requires it.
Every decision is auditable
Complete run traces, model versions, prompts, retrieved sources and human interventions, retained per your policy and exportable on demand.

Start here

Find out where you stand before you talk to us.

Most of the value of an audit is knowing what to ask. Three ways in, in order of how much of your time they cost.

AI Readiness Scorecard0/8

01 / Data

Where does the data an AI system would need actually live?

Playbook / PDF

Build vs. Buy: the 2026 Enterprise AI Playbook

The decision framework we use with clients, and the one we applied to ourselves before building Sanaris, LegalFront and Lightng. Twenty-two pages, including the cost model and the three cases where buying wins.

Session / 30 minutes

The AI Opportunity Audit

Thirty minutes with the people who would build it. No deck, no discovery call in disguise.

  • A scored one-page readiness readout
  • The two workflows with the fastest payback
  • A fixed quote, or an honest “not yet”
Book the audit

Usually within five working days

Start something

Less deck. More shipping.

Tell us what the work looks like today. We’ll tell you honestly whether AI is the right answer for it, and what it would take to put a system in production.

Book a demoTake the scorecard first

Or just watch what we ship

What we built, what broke, what we learned. No sequence.

Talk to our AI