Understand your users' conversations.

Know what your agent is doing for your users, before it becomes a problem.

Built for product managers who want visibility into their AI agent.

Memo · Week of Apr 14–20
Monday, April 20 · 07:128,214 conversations read
01Lead

Users are reconciling invoices with the assistant more than anything else this week.

In 412 conversations ( 29% of volume, up from 18% last week), users asked variations of “how do I match a partial payment to an open invoice.” ev.412  In 83% of these, the assistant recommended a workflow that is not currently supported.

Fig. 01Invoice-reconciliation requests · past seven weeks
412 this week · +1,270% WoW

Prior six weeks averaged 31 conversations. This week: 412.

Supporting evidence · 412 conversations87% semantic similarityhigh confidence

Reads from where your conversations live.

No SDK. No code changes. Point Locus at the trace store and walk away.

Also connects to

At scale no human can read.

Your AI had 94,000 conversations last month. Your team read 212.

This week
10,248 conversations · this week
12 conversations read.
10,236 you didn't get to.
You can only read so many transcripts by hand.
What your analytics see
Clicks this week
Open bookings
13,064
View itinerary
9,088
Start refund
2,556
Change dates
994
Useful. But it doesn't tell you what your users were actually trying to do.
What your users actually sent
Real messages
Can I cancel my Lisbon booking? It's still in the free-cancel window.
Por qué me cancelaron mi reserva de Madrid?
Move my Barcelona dates from June 12–16 to June 19–23.
Change my Paris trip to the first week of June, June 1 through 7, nothing else.
You said a credit would post by Monday. It's Thursday, nothing's there.
Can I bring my cat on the Porto booking next month?
You're saying there's no booking, I have the confirmation email open.
How do I apply my loyalty points at checkout?
Do I need a visa for Portugal if I'm flying from Canada?
This is the third time I've been told to wait for someone.
I need a VAT receipt for a trip my colleague took in March.
EN ESPAÑOL POR FAVOR
Can I cancel my Lisbon booking? It's still in the free-cancel window.
Por qué me cancelaron mi reserva de Madrid?
Move my Barcelona dates from June 12–16 to June 19–23.
Change my Paris trip to the first week of June, June 1 through 7, nothing else.
You said a credit would post by Monday. It's Thursday, nothing's there.
Can I bring my cat on the Porto booking next month?
You're saying there's no booking, I have the confirmation email open.
How do I apply my loyalty points at checkout?
Do I need a visa for Portugal if I'm flying from Canada?
This is the third time I've been told to wait for someone.
I need a VAT receipt for a trip my colleague took in March.
EN ESPAÑOL POR FAVOR

Read. Pattern. Decide. Ship.

How Locus turns a trace into a paragraph your team can act on.

01
Read
02
Pattern
03
Decide
“Cancellation parser shipped Tuesday is mis-reading Flex fares as non-refundable.”
04
Ship
@bookings-teamShip Thursday
Patch cancellation-eligibility parser · 412 guests · $38k ARR

Three things, every morning.

01

Patterns.

Thousands of conversations collapse into a handful of named ones.

Hover a cluster to read one
Every dot is a customer conversation. Every glow is a pattern Locus found without anyone naming it for you.
02

Segments.

Who your users actually are, by how they talk, not their plan tier.

Customer base
10,237 customers
Uncategorized
Customers Locus hasn't confidently placed yet. Usually new or low-volume.
19.1%

Open Locus. See the patterns.

Every morning, a fresh read of what your customers were actually trying to do.

  1. 01
    Open Locus.
    One tab. No queries. Nothing to set up.
  2. 02
    See every pattern your users formed last night.
    Locus clusters thousands of conversations into a handful of named ones.
  3. 03
    Ship the fix before your competitors do.
    Five conversations under the growing pattern. One paragraph to your team.
Today's patterns
Monday · 4 of 8
  1. 01
    Cancellations refused inside free-cancel window
    +1,370% WoW
    412
  2. 02
    Modify a confirmed booking
    6.0% of traffic
    389
  3. 03
    Date-change rephrase loops
    4.7% of traffic
    301
  4. 04
    Travel credit promised, never applied
    3.3% of traffic
    214
4 more in the full memoOpen memo

The missing layer in your AI stack.

Conversations / meaningClicks / numbers
InfrastructureProduct
LLM observability
LangfuseLangSmith
Tells engineers what the model did.
Locus
The missing seat
Tells the PM what the customer was trying to do.
APM / tracing
DatadogOpenTelemetry
Tells SRE the API returned 200.
Product analytics
AmplitudeMixpanelPostHog
Tells growth which button got clicked.
The missing seat is the one that reads.

Every other layer in your stack is pointed at something other than the conversation. Infra watches the API. Evals score responses. Analytics counts the clicks.

Locus is the only one pointed at what your customer was trying to do. At scale, what thousands of them were trying to do together.

Locus does not replace your APM, host your traces, run your evals, or write your prompts. It sits in one specific seat. The one that talks to the person deciding what to ship.
The first time our board review wasn't my PM's guess at what users were doing. It was a paragraph, with evidence, written overnight. That changed the review.
MR
Maya Reyes
VP of Product · Harbor
Built for enterprise buyers

Security a legal team can sign off on.

Locus reads from a trace store that never leaves your infrastructure boundary. The usual B2B checklist, short-form.

SOC 2 Type II
Audited controls across access, change management, and incident response.
GDPR & DPA
Standard contractual clauses and a signable data processing agreement.
No training on your data
Conversations are never used to train Locus or any third-party model.
SSO & SCIM
Okta, Entra, Google. Role-based access, provisioned and deprovisioned in seconds.
Data residency
US or EU region, selected at install. Logs stay in the trace store you already run.
Pricing

Priced on conversations, not seats.

You're already paying for every inference. Locus is a small percentage on top of that, in exchange for the hours of PM time you'd otherwise spend reading transcripts by hand.

Starter
$1,400per month

For a single product team getting a read on conversation quality.

  • Up to 100k conversations / month
  • Read from OTel, Langfuse, Braintrust, LangSmith
  • 3 seats included
  • Email support
Join the waitlist
Team
Most popular
$4,800per month

For multi-product orgs shipping AI in production with dedicated PMs.

  • Up to 1M conversations / month
  • All Starter sources + Datadog
  • Unlimited seats
  • SSO, SCIM, audit log
  • Private Slack channel with the founders
Start a trial
Enterprise
Let's talkannual

For regulated industries, in-VPC deployments, and custom residency.

  • Unlimited conversations
  • SOC 2 Type II, DPA, sub-processor list
  • EU or US data residency
  • In-VPC deployment available
  • Dedicated solutions engineer
Talk to a founder
Connect your traces

Read tomorrow's memo.

No SDK. No instrumentation work. Reads from OpenTelemetry, Langfuse, Braintrust, LangSmith, Datadog, OpenAI, Anthropic.

Your first memo ships in 24 hours.