Release notes

From a spreadsheet to a sold car, all from the phone

Grove 1.1.0 finishes what a seller and a buyer start: a car from the week's de-fleet spreadsheet to live, an offer, a counter, and every step of the deal to pickup, a card per step on each side, with no Cypress Manager needed. Cypress adds what that takes (API 1.10.0 and 1.11.0): paying the commission by card from Grove, each party seeing only its own signing links, the year ahead on every value, what holding a car costs, and pricing rules that list only ready cars.

Grove: your team, your Cypress Manager and Cedar, on your phone

Until now Cypress was one app, made for a browser on a desk. The people who sell and buy fleet cars spend their day in a yard, a truck or a lot, and we expect about nine in ten of them to reach Cypress from a phone. Grove is Cypress's second app: a phone-first team app where an organization's people, its Cypress Manager and Cedar work in one place, and where Cypress's decisions come to a person one at a time, with the work done. Grove keeps no record of its own about cars or deals: it reads and writes Cypress only through the public API, as the person signed in, so everything it does a partner's app could do too. Making that possible grew the API to 1.9.0: every member can now sign in to a connected app, a Cypress Manager can act for an organization from one, and offers and closings come with the facts worked out. It is the first time Cypress is two apps, which is why this is a major version. Grove 1.0.0 and Huddle 1.0.0, the conversations service both apps use, ship with it; each now has its own version and changelog.

Answers given while generating can be changed or cleared

A completed step is reopened only when a Cypress Manager says so

The gate pass says where to pick the car up

Your inventory, kept in step

A fleet customer's first API integration asked for two things the API made hard. Their feed has no photos, only a link to each unit's Manheim condition report, and the API had nowhere to put it; the report endpoint we did have was documented as a CARFAX/AutoCheck reader, which it is not. And they wanted to push their whole inventory and have Cypress take off whatever was no longer in it. Both now work. Along the way the archiving rules changed for everyone (owner and CEO): a seller's archive waits for a Cypress Manager only when a buyer is involved.

Invoices your AP can book

Some fleet customers' accounts payable match every invoice to a purchase order, often one per branch, and book invoices by branch or by car. Cypress invoices now carry what they need: a PO on each location that flows through Deal Setup onto the invoice, an invoice line written the way the customer's AP reads it, and, for ACH and bank-transfer customers, one invoice per close, per location, per PO or per vehicle. Billing emails can go to several addresses.

Platform activity

Cypress now knows who uses it and what they do. Every visit, sign-in and act is recorded, with who did it (the customer's own people, Cypress, or an agent) and how (by hand, in bulk, with Cedar, or through an agent), and a new staff page answers the CEO's questions: active users by side, the funnels from first visit to sale, every buyer and seller action, each organization person by person, the Cypress team's work, and who comes back. Actions carry their whole history from the first deploy; visits and sign-ins start today. The same figures reach the API, MCP and Cedar. PostHog had been losing every signed-out visitor since July 10; fixed.

Pay by invoice

A third way to pay Cypress, for a customer whose accounts payable pays invoices and never lets a vendor pull money: **bank transfer**. The same terms as ACH, one invoice a cycle with a due date, an account number of its own, and payments matched automatically. The billing screens are rebuilt around how an organization pays, Lago is gone for good, and the ACH cycle's long-standing faults are fixed on the way.

The grand MCP release: Cypress is agent-ready

Cypress opens up to software. Everything a person does on the app's screens can now be done by an AI agent or a partner's app, through MCP or the REST API, under the same rules, with the decisions that belong to a person handed back to one. Agents get thirteen core tools instead of four hundred, a sandbox of their own that resets itself, and a documentation section written for them. The Integrations API goes to 1.5.0. Includes every fix prepared as 3.4.3.

Undo means undo, and a live deal is never archived by surprise

A storefront takes offers on every car it shows

Plan what you buy, with Sequoia on both sides

Buying and selling become one planning tool, and the loop between them closes. A dealer plans the quarter's purchases the way a seller plans its sales, on the same calendar. Sequoia reads the market for both sides and records every suggestion with its reasons. Both plans follow one quarter bar, keep Sequoia's rail beside every tab, and carry a Market tab of what the plans, searches, saves and offers on Cypress say is coming. The plan follows each side into the Marketplace, Buyer Hub and Seller Hub, and Cypress staff get a Cypress desk on both pages to broker the fit. Ships together with 3.3.1.

Buyers see how a seller decides, never where

Seller terms buyers accept, and closing paperwork that looks like the seller's own

Sellers can now publish their own terms, such as Fox Rent A Car's Vehicle Sale Return & Arbitration Policy, and every buyer must accept them before making an offer or buying. The accepted version is recorded on the offer and printed, word for word, on a Seller Policy the buyer e-signs at pickup. Fox's own Bill of Sale, its Manheim variant and its Gate Release are rebuilt as templates, and a deal can switch between Bills of Sale in Deal Setup.

One email per bulk action, and campaigns that never mail a sold car

Three hotfixes ship together. Bulk actions no longer flood inboxes: on October 5 a bulk launch of 121 Fox Rent A Car vehicles sent 968 "Vehicle Listed" emails, 121 to each of Fox's seven Fleet Managers and its account manager, with no VIN to tell the cars apart. Vehicle notifications now arrive as one email per person and type, with a table of the vehicles. Marketing campaigns stop mailing vehicles that are sold or in a deal, and pause themselves when nothing is left to sell. Notifications meant for Cypress, which looked for Cypress staff inside the seller's own organization and so reached nobody at most accounts, now copy the account's assigned account manager.

Planning over the API, and marketing mail straight to Resend

Two things ship together. The partner API catches up with Planning, which it should have done in 3.0.0: in three contract steps (API 1.1.0, 1.2.0, 1.3.0) it stops treating planning stock as Sell stock, tells an integration what its key may do, and then lets a fleet's own systems run Planning end to end. And marketing mail — promotion campaigns, CRM campaigns and one-off outreach — now goes to Resend directly instead of through Novu, so a busy moment no longer drops sends silently.

🧭 Planning: deciding what to sell, and when, before it's sold

Until now Cypress started at the moment a fleet had already decided to sell a car. Everything before that — which vehicles leave this quarter, in what order, against what target, and whether holding one more month earns or burns money — happened in spreadsheets, and arrived in Sell as a pile of drafts with no reasons attached. This release adds that step to the platform. It is the first time the product is two stages instead of one, which is why it is a major version.

🗂️ One license per state, and a counter Cypress can answer

Two fixes to places where the platform held a single slot for something the business has more than one of.

⚡ The send that stopped taking the site down

Twice this month a campaign send took production down, and the number that explained it was not in the database. It was in the mail: a corporate link scanner opens every URL in an email the moment it arrives, so 2,100 recipients became thousands of clicks in a burst, each one a tracking write and a full page render. The first fix bounded how many of those could touch the database at once. It stopped the errors and left the slowness, because every recipient of a send shares one row and the four lanes still queued behind its lock.

🤝 The closing that runs itself

Nine steps became six, and the three that went away were not simplifications — they were places the platform had been asking a person to tell it something it already knew. A buyer attesting they had sent money, a Cypress Manager reading a VIN back off a document Cypress had just generated from that VIN, a buyer confirming a title arrived on a step of its own. What is left is the work that genuinely needs a human: signing, confirming money landed, releasing a vehicle, collecting it.

🖼️ Seeing the email before it sends

A promotion went out with a square logo as its hero. In the inbox that logo is not a logo, it is a 532px band of white above the first car — because the email is the only surface that renders the image at the shape it was given, and nothing in the product had ever shown that shape. The campaign screen's preview drew the author's prose and stopped there; the hero was a 144×80 thumbnail with `object-fit: cover`, which hides aspect ratio by definition. The one place the framing was visible was the sent email.

💳 The number in the email is the number on the card

2.9.3 gave an invoice three ways to get smaller. This release makes the emails say so — found by testing the flow rather than reading it.

🧾 Money that adds up, and mail that arrives

Two halves, both about a number on a screen being wrong in a way nobody inside the product could see. An invoice settled by a mix of credit note, account credit and card had three partial records and no document tying them together. And the biggest customer on the platform had stopped receiving campaigns seven weeks earlier, because a mailbox that was briefly full got treated as a mailbox that was gone.

🩹 The e-signature service was never unreachable

The Bill of Sale that started 2.9.1 still would not send for signature, and production still said **"the e-signature service was unreachable"**. It was reachable the whole time. Every failed attempt had created the signing template it needed, and then thrown it away.

✍️ The paperwork signs itself: one authority, one card, one honest state

A production Bill of Sale could not be forwarded to its buyer. Approving it asked the Cypress manager *"how does the buyer sign?"* — about a document generated from a template that had already answered — and answering returned **Internal server error**, five times in five minutes. Pulling that thread ran the length of the closing-document platform: who decides how paper is signed, when the buyer is allowed to hear about it, and whether the screen is telling the truth about either.

🧭 The Customer 360: Performance, Engagement, and Sakura's read on both

The platform could say what happened — sends, sales, offers — but not who anyone **is**. A company's browsing, purchases and CRM record lived on three screens that never met. This release rebuilds analytics around one spine: **Engagement** is the demand half, **Performance** is the money half, the **CRM** acts on both — and every screen carries the same **As buyers / As sellers** lenses, because one company can be hot on one side of the market and cold on the other.

🎯 Reached means we actually sent it

The Marketing screen said **4,351 people reached** for a campaign whose entire book is 2,151. Three separate mistakes were stacked inside that one number, and every rate on the screen divided by it — so every rate was wrong too. This release makes one word mean one thing, and stops a public page crashing when its own API blinks.

📊 Numbers that mean what they say, and the addresses you cannot reach

A 2,095-recipient campaign — the first send in this system's life to reach a real audience — took production down for twenty-five minutes and then reported its own results wrongly. This release is the remediation and the reckoning: every engagement figure the marketing screens showed was measuring something other than what its label claimed, and the suppression list the mail system had been enforcing for months was visible nowhere in the product.

🌲 Cedar learns the rest of the app, and express offers stop hiding

Cedar's brain had kept pace with the vehicle lifecycle; its **map of the product** had not. Fifteen releases added CRM, Mission Control, Marketing, Promotions, Storefronts, Billing and Analytics to the navigation, and not one of them touched the files that tell Cedar those places exist — so the copilot could not name them, reach them, or automate them, and nothing in the build ever objected. Alongside that, the express offers 2.8.5 made visible to sellers were still invisible to everyone else.

📊 Sheets you can sort, offers the seller can see

A patch on the vertical 2.8.4 opened. Three small things, each of which was costing somebody real work: a spreadsheet that could not be sorted by year, an offer the seller was never told about, and a marketplace that opened as a wall of photographs when the buyer wanted a table.

🌎 Regions, headline cars, and the offer that needs no account

One initiative, end to end: promote cars by region, lead the campaign with the cars you chose, and let the dealer who opens the email answer with an offer — without making them register first. The marketing email finally closes its own loop: region-targeted promotion → featured cars up top → reply or one-click express offer → a deal in the CRM, whether or not the buyer has ever logged in.

🖋️ The rest of the paperwork, signed either way

2.8.2 fixed how the Bill of Sale gets signed. It turned out to be the only document that could be signed at all: everything else — a gate pass, a title, any slot on the checklist — refused an uploaded file outright with *"Only generated documents carry signature areas"*. A seller who uploaded their own paperwork had no route through the closing except to handle it entirely off-platform and tell nobody.

✍️ The paperwork, and who signs it

A Sentry alert on a field that could not be filled in turned into a week's worth of the closing's paperwork being wrong in ways nobody could correct. A custom field defined after a deal started could never be answered. A seller address fixed in the profile never reached the document. An uploaded photograph was accepted and then rejected three clicks later by DocuSeal, as a bare "Internal server error" that Sentry never saw. And the signature it did place went at a fixed spot on page one, over whatever happened to be printed there.

📇 A pipeline with the whole book in it

The CRM shipped after hundreds of cars had already been offered on and sold, so the board showed fifteen deals against three hundred offers — and could not have shown the rest even if they had existed, because two separate filters made the settled lanes unreachable.

🤝 A deal closed elsewhere, and a bill that waits its turn

Two things had no home. A sale agreed outside Cypress could only be finished by pretending it happened here — the Bill of Sale step's one exit raises an e-signature request for a document signed in ink weeks ago. And the commission scheme, correctable only until the invoice exists, was billed within the hour of acceptance, before anyone had looked at the deal.

🔒 A rule cannot un-sell a car

An offer arrived on five vehicles and the platform recorded it correctly. The next morning, lowering their listing price re-ran a listing rule over all five and quietly put them back on the market with the offers still open. Two independent faults had to line up: a trigger that fired on a field nobody had changed, and an action that would overwrite anything.

📣 A promotion that says what you want it to say

A filter-criteria promotion turned out to be invisible to everything except its own page: the campaign screen called it empty, the offer sheet built no rows, and the email carried no cars. Fixing that exposed the next thing — a promotion email had no words of its own to change — so it has them now, and the link box stopped demanding that anyone know what a URI scheme is.

📨 Messages you can reuse, sheets you can send

Everything in this release is about the moment a spreadsheet or an email leaves the building. Three exports where there was one, a promotion that can carry its own vehicle list, delivery reporting that lands on the right recipient, and — new — a library of reusable messages with a rich-text designer.

✍️ One-off CRM outreach sends again

An audit of the three Novu environments after 2.6.1 found `crm.outreach` sitting in **ERROR** in all of them — local, staging and production alike.

🩹 The audience finds its people, and the flow editor opens again

Everything 2.6.0 shipped, put in front of a real book. One release-stopper and four places where the screen was quietly the last to know.

🗺️ The book knows where it is, and the send says what happened

Living with the v2.5.0 import at real scale — ~2,900 rows in one file — surfaced what a book of that size actually needs. A location you can filter on. A grid that groups. A way to act on four hundred records at once. Flows that fire on people and companies, not only cars. And a campaign screen that can tell "nobody opened it" apart from "it never went out." Five workstreams, one release.

Lago is decommissioned, and its resources come home

v2.2.0 moved invoicing to Stripe and switched Lago off behind `LAGO_ENABLED=false` — but the machinery kept running: six Railway services (API, worker, clock, PDF, front, and a dedicated Postgres), six local containers, and an activation path that still minted customers into a ledger nothing reads. This release retires all of it and reclaims the compute.

📇 The book fills itself: CRM import arrives

Prospect lists stop being spreadsheets someone means to type in one day. **CRM → Import** takes the file as it is — a conference dealer list, another CRM's export, a colleague's collection of columns — and turns it into contacts, businesses and deals in one pass, matched against the book so nothing you already know is ever duplicated. The first real cargo through it: ~2,900 rows of dealerships that landed as 1,967 contacts and 2,352 businesses, found afterwards with a single filter. And because a 3,000-record book is a different animal from a 30-record one, this release also makes the CRM lists show *all* of themselves instead of a silent first page.

Every seller archive is reviewed, and the review queue stops hiding

2.4.1's audit of the archive machinery surfaced a policy contradiction: the fleet panel staged a seller's archive for CM review while Seller Hub archived the same inventory immediately, so whether a removal got reviewed depended on which page the seller happened to be on. This release picks a side — review, always — and then makes sure the resulting queue is impossible to miss, because a vehicle with a pending archive request is deliberately untouched: still listed, still able to take offers, until a CM decides.

Locking the side doors: the satellite apps get auth, sellers get their tools, the database gets a ceiling

A hotfix with one headline and a common thread. The headline: ARVIS (`inspect.*`) and Orbis (`data.*`) had been deployed with no authentication at all — every page and API answered anyone on the public internet, including the new-inspection page, which listed every organization, fleet, vehicle and VIN on the platform. The thread: a dealership's own ADMIN kept being mistaken for platform staff by the UI (handed archive-outcome modals the server would only refuse) while being denied the seller tools they actually own. Both are the same confusion about who someone is, resolved in opposite directions.

Delegation: what Cypress may do for you is now something you grant

Cypress managers have always operated inside customer accounts — the question was on whose authority. The honest answer was "a role string": a checkbox labelled *Cypress Managed* that gated nothing, an offer panel that decided who may accept by comparing the offer to the floor price, a `COUNTER` action with no role check at all, and — worst — two vehicle routes that trusted an `actorRole` field in the request body, so any authenticated user could claim to be a Fleet Manager, perform FM-only transitions including the floor-price consent, and have the audit record the claimed role as fact.

The quiet failures: spend, silence, and errors that named the wrong thing

A day of production archaeology rather than new surface. Every item below was already happening; the work was noticing. Several had been running for weeks behind a log line that read correctly, which is the thread connecting them: a guard in only one of nine places, a fallback that turned a typo into a connection error, an audit write that failed after the request had already succeeded.

CRM and Marketing: the prospect side of the business, in the platform

Until today the commercial team ran the prospect side of this business outside the platform. Contacts lived in a spreadsheet, the list of who to mail was rebuilt by hand each time, the mail itself went from a personal inbox, and what came back — a reply, a call, a "not this quarter" — was remembered by whoever sent it. Cypress could tell you everything about a vehicle and almost nothing about the person who might buy it: interest arrived as an engagement event, became a chart, and stopped.

Billing moves to Stripe, and stops guessing

Lago was quietly trying to charge every invoice the moment it finalized, using its own copy of each customer — a second collection system nobody had asked for, failing on every attempt, and invisible to us because its payments carried none of our metadata. That is the thread this release pulls. Stripe becomes the accounting ledger, Cypress keeps collecting, and the invoice a seller downloads is the same document either way.

Cedar moves into the docs

The manual stops being a place you read and becomes a place you can ask. The platform's own assistant now lives in the documentation — not a lookalike widget, the same controls the app mounts — grounded in these chapters and able to walk you to them. And every chapter opens with a plain-English summary for the reader who has ninety seconds, not an afternoon.

The platform gets its manual

/docs grows from developer documentation into the full story of the platform: a user manual for every seat at the table, sitting beside the developer docs it used to be.

Payments can't fail quietly, activation can't go unnoticed

A dealer finished onboarding, a manager activated the org — and the dealer still saw "Limited access" and couldn't pay a single invoice. The trail led back to the 8/11 Stripe account cutover: the new live account never had ACH activated, and a checkout that asked for `us_bank_account` was rejected whole, cards going down with it. Every invoice payment since the cutover had silently failed. The capability is enabled now; this hotfix makes sure neither half of that morning can repeat.

The API joins the automation engine

The last parity gap from the partner test drive: API pushes now drive the same automations everything else does — and flows gain a say in who gets to trigger them.

First partner test drive, three rough edges

The partner's very first real push through the new API surfaced exactly the feedback a 2.0.0 needs: a vehicle created cleanly, but the embedded listing vanished without a sound, the webhook arrived in an unexpected vocabulary, and the Bruno kit refused to import. All three are the difference between an API that works and one that explains itself.

Cypress becomes a platform other systems build on

A partner asked to push inventory into Cypress the way sellers integrate with Manheim. The answer grew into the largest single release since launch: a public machine-to-machine API with two dialects, signed webhooks that narrate the whole remarketing process, self-service integration management for every business org, a public developer portal with living documentation, and a complete testing kit. The 2.0.0 mark is deliberate — Cypress is no longer only an application people log into; it is a platform other systems integrate with.

The closing stops being a black box

A test sale surfaced three ways the closing kept its work to itself: an error message that read `[object Object]`, a "Last Update" column frozen on something that happened 67 days ago, and a vehicle timeline that went silent for the entire sale. Fixing the reporting turned into a pass on the moments a deal is meant to speak up — and on the small dead ends that make a seller leave the page to finish a sentence.

Chains get their own trigger, and priorities get a save button

A day of living with v1.9.6's automation ordering surfaced the rough edges: chained flows had no honest trigger to wear, priority edits saved silently with no feedback, and the chain relationships were invisible unless you already knew where to look.

The VIN sells the car, and flows stop racing in the dark

The business was blunt: without the VIN on the page, buyers don't buy. And in Automations, two things had quietly become unworkable — the switch node's branches couldn't be wired to anything until a page reload, and flows sharing a trigger ran in whatever order the database felt like, with no way to even see which flow fires on what.

The commission is a gate, not a formality

A test sale produced a commission invoice that was born "Paid" — no card was ever charged — and the closing marched on regardless. Two independent defects lined up: the closing completed its fee step at invoice *creation*, so payment was never actually a gate; and Lago auto-marks a one-off invoice "succeeded" when the customer has no payment provider linked, then delivers that verdict by webhook after the fact. Fixing the pair, and then walking the same closing as the buyer, the seller and the manager, turned into a broader pass on who must pay before what — and who gets to see whose money.

Notifications that go somewhere

A commission invoice email arrived with a "View in billing" button that went nowhere: its link was `/billing`, a path with no host, which a mail client can't open. An audit of the notification system showed this was the rule, not the exception — every notification travels through one worker that passes its link to the email template verbatim, and roughly sixty call sites across billing, closing, offers and the fleet were handing it app-relative paths. The same audit explained why in-app notifications read like announcements instead of actions: the payload always carried a destination, but the in-app channel was never told about it.

Who may do what: roles said out loud

A Cypress Manager selected three DRAFT vehicles and was told only Fleet Managers could submit them — while the same person could submit them one at a time. Pulling that thread surfaced a pattern: nineteen routes tested "the role is CYPRESS_MANAGER or ADMIN" and called the result *platform staff*, but ADMIN is an org-scoped role — a customer organization's admin passed every one of those gates. This release makes the permission model say what it means, and makes several fleet actions genuinely seller self-service.

Closing documents: designed, generated, signed

Closings used to run on two hardcoded upload slots — a Bill of Sale and a gate pass, each a PDF somebody produced elsewhere. This release replaces that with a document platform: Cypress and each organization design their closing paperwork on a canvas, the deal's facts live in one place both sides confirm, documents generate themselves from those facts, signatures are collected in the declared order — electronically through the platform or in ink at the yard — and the checklist knows the difference between where a document is *born* and the step by which it must be *fully signed*.

Inventory reports by email, and a vehicle that can sell twice

Two halves. Automation flows have always acted on one vehicle at a time; this adds the nodes a CM needs to act on a *set* — collect a channel-filtered slice of inventory, put it in a spreadsheet, email it out on a schedule. And a vehicle whose first sale fell through could never be sold again through Post-Sale: the second deal was rejected by a database constraint, the error was swallowed at every call site, and the buyer was told "Purchase Confirmed!" for a sale that existed nowhere.

Cypress was its own biggest customer

The retention report named **Cypress** as a buyer of 76 vehicles. It never bought any of them.

Hotfix: the Analytics page crashed at every stage

One definition of a sale, across the whole dashboard

1.7.5 put a stage switch on two tabs. Two tabs on one page could therefore report different numbers, which is the thing the switch exists to prevent.

Sale markets, release history, and a sales time-lapse

A pitch-deck slide animating every sale onto a US map turned out to be two datasets stitched together — and chasing its data found a $2.03M gap between the operator's sales book and what the platform reports.

Vehicle fields that can be cleared, and engine fields that save at all

The sibling defects 1.7.3 flagged and deferred. One of them turned out to be worse than described.

Vehicle identity: trim recovered from the model column, and a specifications card with something in it

Two halves of the same complaint — the platform knew less about a vehicle than it could have. Sellers were handing us the trim inside the model column and we threw it away; the VIN decode was handing us engine and body data and we never wrote it down.

Hotfix: three broken redirects

Consumer accounts made usable, seller-owned marketing, and the first production hardening of the remarketing vertical

v1.7.0 built the remarketing vertical; 1.7.1 is what it took to put it in front of real people. A consumer can now sign up with Google or a password that actually works, confirm their address once at the account level, and manage the sellers they hear from. Sellers get their own promotions, campaigns and analytics — scoped to their own storefront rather than the platform's. And three authorization defects found on the way to prod are closed.

Remarketing OS: Sales Channels, Direct Links, Storefronts & Consumer Accounts

The remarketing vertical. Inventory stops being "listed or not" and gains **sales channels** — the public Marketplace, **Direct Links** (private share links), and a seller's own white-labelled **Storefront** — each an opt-in grant with its own commission scheme, its own attribution, and its own fee at closing. Sellers choose channels per vehicle or in bulk; buyers arriving through any door are told which one they came through; and a storefront can now acquire a **consumer** directly, so an individual can register, subscribe, and make an offer without ever being a dealership.

Marketing landing mobile nav

Hero ad under reduced motion

Retention axis & dormant clients

Client Cohort Retention

Engagement Analytics, Promotion Marketing & Admin Org Parity

An analytics-and-marketing release. A full **Engagement Analytics** suite lands on both the promotion Analytics tab and the vehicle Market History — KPIs, a conversion funnel, engagement/activity heatmaps, a real geographic map, audience breakdowns, visitor and per-vehicle grids, a **Source → Vehicle → Outcome → Sold Sankey**, and a **Sakura promotion insight**. Promotion **marketing campaigns** gain a full audience model (filter / cherry-pick / paste-an-email-list), tracking, and sends. Cypress Managers get real **org parity** — upload documents and fully manage a specific org's pricing rules on behalf of a business org — and the dashboard adopts a **shared header/tab UI convention**.

Market History Intelligence, Sakura Market Summary & Buyer-Facing Seller Anonymization

A seller-analytics and buyer-privacy release. The **Market History** tab is rebuilt into a real engagement-intelligence view — Hot/Warm/Cold status, the buyer network as a baseline, and each seller-side update correlated with its impact on engagement — threaded through the fleet grid and a branded PDF report, and narrated by a new **Sakura Market Summary**. On the buyer side, the Marketplace and Buyer Hub are consolidated onto one vehicle-detail component and **seller identity is anonymized** across all buyer-facing surfaces.

Public Marketplace, Curated Promotions, Marketing site & Sakura Vehicle Insight

A seller-acquisition and public-facing release: an unauthenticated **Public Marketplace** lets prospective buyers browse and drill into real inventory before registering, **CM Curated Promotions** replaces the old Campaign scaffolding with a real promotion builder (cherry-pick or live-filter vehicle sets, AI hero images, scheduling), a new **marketing home page** replaces the bare site root, and vehicle detail pages gain seller credibility, condition reporting, related vehicles, and an AI-written **Sakura Vehicle Insight**.

Marketplace seller links & automation trigger fix

Email normalization

User-admin parity

Sales reporting & removed-sale visibility

A buyer-side release: the **Marketplace** gets a ground-up redesign with an AI-assisted search, smarter filters, Sakura recommendations, and a real **alerts engine** (saved searches that watch inventory, notify, digest, and auto-save). Cedar's Copilot reaches the **Buyer/Seller hubs and offers** — it can now read offer status and accept/counter/reject behind the confirm-rail. Pipeline and Today metrics are re-grounded on **wholesale GMV** (the figure Cedar already cites) with a separate retail KPI.

Cedar becomes a true co-pilot: a new **Copilot workspace** in the app and a full **Telegram channel** — both driven by the same agent, tools, and confirm-rail, so anything Cedar can do, it can do by text, by voice, or from your phone.