Skip to content
SOC 2 Type II · GDPR ready

Every answer, with its receipts.

Halyard drafts replies from your documentation, your tickets and your product data — and shows the passage behind every sentence. Your team approves the tone once and stops rewriting the same answer.

Free for 14 days · No card required · 12-minute setup

Customer teams already answering with Halyard

  • Northwind
  • Everline
  • Blackford
  • Quanta
  • Riverstone
  • Palladio

The gap

A fluent answer and a correct one look identical.

That is the whole problem with putting a model in front of a customer, and it is not a prompt-engineering problem.

Where it breaks today

The assistant invents the specific bit

It gets the shape of the answer right and the refund window wrong. The parts a customer will act on are exactly the parts a model is least reliable about.

What Halyard does instead

Nothing ships without a source

Every factual sentence is matched back to a passage in your own content. Anything Halyard cannot ground is flagged before a human sees it, not after a customer does.

Where it breaks today

Your knowledge is six years of scattered writing

The real answer is split across a help centre, a Notion page, last quarter's release notes and a reply someone wrote at 11pm in a ticket.

What Halyard does instead

One index over all of it

Connect the sources once. Halyard reads across every one of them in a single pass, respects the permissions each system already enforces, and re-indexes as they change.

Where it breaks today

Nobody can explain a wrong answer

When a reply goes out with a bad number, the team needs to know which document said it. Most tools can only show you the prompt.

What Halyard does instead

Every answer is fully traced

Retrieved passages, grounding score, the model that wrote it and the editor who approved it — recorded per answer, searchable, exportable.

The workspace

Open any answer and read where it came from.

Which documents were retrieved, how each sentence scored against them, what the reviewer changed and what finally went out — the whole path, per answer.

01

Sentence-level citations

Each claim links to the passage that supports it, not to the document that contains it.

02

Grounding score

One number for how much of a draft is supported. Set the floor; the rest goes to review.

03

Cost per resolution

Token spend attributed to the conversation it closed, not to the month.

Capabilities

Built for the team whose name is on the reply.

Everything below ships on every plan. The Enterprise tier adds controls, not features.

Grounding guardrails

Write what Halyard may and may not say in plain language. It becomes a hard check that runs on every draft before anyone reads it.

Compiled to

  • Minimum grounding
  • Banned claims
  • Required disclaimers
  • Escalation triggers

Review before send

Anything under your grounding floor, or touching a topic you flagged, waits for a person. Everything else goes out on its own.

Answer history

Re-run any past question against updated documentation and diff what would be said differently today.

Bring your own model

Route per queue to any provider, or to a model inside your own VPC. Swap without re-indexing a single source.

Runs where you want

  • Hosted
  • Self-hosted
  • In your VPC
  • Per-queue routing

Sixty native connectors

Index your helpdesk, wiki, CRM and release notes. Anything missing takes an afternoon over the SDK.

Reads from

  • Deskline
  • Clarity
  • Kettle
  • +57 more

Enterprise controls

SAML SSO, SCIM provisioning, regional data residency, customer-managed keys and a signed DPA.

Getting started

Answering by Thursday, not next quarter.

No professional services engagement. No content migration project.

  1. 1

    Connect your sources

    OAuth into the helpdesk and the wiki you already pay for. Halyard inherits the permissions each of them enforces.

  2. 2

    Set the floor

    Pick the grounding score below which nothing sends, and the topics that always go to a person. Two decisions, not a configuration project.

  3. 3

    Run it in draft

    Halyard writes but does not send. Compare a week of its drafts against what your team actually replied.

  4. 4

    Turn on sending

    Promote one queue at a time, and drop back to draft just as fast if the satisfaction numbers move the wrong way.

Connected

It reads from wherever the answer already lives.

Sixty native connectors, a typed SDK for everything else, and no per-source pricing.

DesklineSupport
LedgerlyBilling
ClarityCRM
HarborWarehouse
RelayMessaging
BeaconAnalytics
TallyFinance
SiftRisk
CargoLogistics
PrismIdentity
KettleData
MeridianPlanning

In production

The numbers our customers report back.

  • 6.8M

    Answers drafted each month

    Across 1,400 workspaces

  • 68%

    Of conversations closed first-touch

    Median, first 90 days

  • 12 min

    To a first grounded answer

    From signup, median

  • 99.98%

    Service uptime

    Trailing twelve months

Proof

Teams that stopped rewriting the same reply.

Northwind

We had a help centre nobody read and a support team retyping it forty times a day. Halyard turned the help centre into the answer, and the writing we did last year finally started paying for itself.

First-touch up 22 points

Imogen TraoréVP Customer Experience, Northwind
Blackford

The citation panel is what got this past legal. They could see, per sentence, which of our published documents the answer rested on. No other tool we looked at could produce that.

Approved in one review

Daniel OkonkwoHead of Risk, Blackford
Everline

Our macros were four hundred snippets that all disagreed slightly. This is one index and a grounding floor, and the answers got more consistent, not less.

400 macros retired

Sara LindqvistDirector of Support, Everline

Pricing

Priced on answers, not on seats you're afraid to add.

Invite the whole team on every plan. You only pay for drafts that get used.

Billing period
Save 20%

Starter

For a first team putting one queue on autopilot.

$39$49/month, billed yearly

You save $120 a year

Start free
  • Unlimited teammates
  • 2,000 answers each month
  • 10 connected sources
  • Draft mode and citation history
  • Community support

Growth

Most teams start here

For support and success teams answering at volume.

$159$199/month, billed yearly

You save $480 a year

Start free
  • Everything in Starter
  • 40,000 answers each month
  • All 60 connectors, plus the SDK
  • Review queues and grounding guardrails
  • Bring your own model
  • Priority support, 4-hour response

Enterprise

For regulated teams with their own security bar.

Custom

Talk to sales
  • Everything in Growth
  • Unlimited answers, committed pricing
  • SAML SSO and SCIM provisioning
  • Regional residency and customer-managed keys
  • Private model deployment in your VPC
  • Named engineer and 99.9% SLA

All prices in USD. Overage billed at $0.003 per answer — never a surprise invoice.

Questions

The things people ask before signing up.

Still unsure? A support engineer answers, not a bot.

Talk to a human

It says so, and the draft goes to a person instead of to the customer. Every sentence is scored against the passages that were retrieved for it; if the score falls below the floor you set, the answer never sends on its own. Silence is a supported outcome — inventing an answer is not.

Point it at what you already wrote.

Connect one source, run a week in draft mode, and read the diff against your team's replies. If it isn't better, you've lost an afternoon.

Free for 14 days · No card required

Live demo of Vantis · see all five brands