Canopy

Diligence · R3

Original document · R3 bodyUpdated 17 Sep 2026 · plain-language EO copy

Startup plan

Canopy is developing the operating system for aerial field work. Narrow commercial experiment plus a separately gated R&D track. Own-field Bulgaria test ~1,000 acres (~405 ha).

Research provenance. This page is an HTML copy of the founding operating memo. Informal fleet slang in the historical body is retained for attribution. Investor-facing product surfaces use professional industry language.

Updated: 2026-09-17 · plain-language EO copy

Earlier body stamp retained for provenance: Updated: 2026-09-15 · R3. R3 merges the current commercial experiment and separately gated R&D track into this founding plan. First/only claims and automatic-compliance promises are replaced with tested acceptance requirements. Historical August 2026 body text is retained and qualified, not silently deleted.

The operating system for aerial field work.

Date: 16 August 2026

Owner: AT / Velocity EU

Audience: founders. Not a partner deck.

Companion: research/10-sat-fleet-viability.md (sourced sat verdict). Fleet fork: research/11-platform-vs-robotaxi.md. Ops / crop × treatment: research/12-ops-pricing-crops.md. Designed end-state: product/option-c-hybrid-fleet.md. Operator ingest (existing crews’ path + receipt): product/operator-fleet-ingest.md. Current commercial/R&D memos: research/releases/2026-09-15/multisector-strategy-r2.md, research/releases/2026-09-15/autonomy-integration-r2-1.md. Abbreviations: docs/glossary.md. Product pages under product/.

Facts are sourced. Dollars that are not published are labelled HYPOTHESIS or ask. Unpublished prices, take rate, seed size and valuation remain open. No open grant or award is claimed.

Current direction (R3 · 15 September 2026)

Two separately gated tracks. Do not fund them as one simultaneous commitment.

  1. Commercial experiment. Recurring property / construction documentation for a paying portfolio buyer (Capture / Portfolio / Assured routes). Test homeowner and promotional media through agents and agencies as a second channel. Agriculture remains a specialist vertical. The US, EU and UK are not greenfield. Prove a buyer-specific advantage.
  2. R&D experiment. Canopy Link is a hypothesis: two named, supported ecosystems; evidence quality; one existing payload before proprietary hardware. Supervised site automation only after permissions, technical evidence, a sponsor and financing. No open-call or grant claim. The reviewed Innovate UK dual-use aviation/autonomy call closed 3 June 2026. Reviewed Eurostars Call 11 closed 10 September 2026. Keep unawarded grants out of cash.

Own-field Bulgaria test (~1,000 acres / ~405 ha founder-stated base): request → accept → live map → BFSA-shaped receipt. Own-land can test UX, receipt fields and ferry notes. It cannot prove marketplace liquidity or paid arm’s-length demand.

Ingest rule: a phone or iPad in a truck reports crew/device position, not the aircraft flight path or applied coverage. A telemetry export is not automatically a legally accepted pesticide application record.

Compliance rule: software can assemble candidate BFSA / EASA fields. Acceptance is a tested human/directorate decision, not an automatic legal receipt.

What this company is

The operating system for aerial field work. Canopy is that proposed OS — job booking and matching as workflow, not a uniqueness claim. Request/accept may use a familiar ride-hailing UX metaphor; that is secondary to the product definition.

A landowner books a job on a mapped plot. An operator flies it. The customer watches live if a labelled source is available. The receipt is assembled from source evidence and still needs acceptance. Satellite and weather support maps sit on the same map as the job. Civil protection may use the same Operating System (OS) only after a named sponsor, lawful use and an accepted data boundary. Shared software does not confer public-safety or military qualification.

We do not sell satellites.

We do not sell drones in v1; Option C may lease/host docks later. We still do not become an Original Equipment Manufacturer (OEM).

We do not sell pesticide.

We sell a completed job.

v1 is still the platform. The designed end-state is Option C (Tesla hybrid). That is a schema and a later dock test. It is not a walk-back of the first product.

The physics of the data layer

Unused satellite coverage over land with no booked work is wasted cost.

Sentinel-2 access is already free (~10 m, ~5-day, Copernicus Data Space Ecosystem (CDSE) — APIs). Prefer free Copernicus / Sentinel imagery when it is good enough. Do not buy commercial satellite imagery only to replace free public imagery. Free access is not zero all-in processing, storage or delivery cost.

Planet and Balkan exist. Buy them on booked hectares only. Cost sits in the job, not in a sat P&L. Every extra paying hectare on the same contract drops $/ha. That is how the cost of this data actually falls.

Starlink works because every sat is full. A private agri fleet over Bulgaria (BG) is empty most of the day. A later MAF land-use series exists; do not headline it until the primary table, year and denominator are reconfirmed. Census 2020 (No406) remains the holdings-shape source. We need paying fields, not the country.

Owning birds before the contract is full is the traditional move. We do not make it.

Optical dies in cloud. Reliability is Sentinel-1 / a drone, not more optical sats. The drone is the product sensor. Satellite and weather maps are support layers: they help identify where work may be needed and check results after a flight.

EnduroSat Space Service marketing slide (234 sats / $344M / $1/GB vs $3,000/GB) is a VENDOR SLIDE, not a Canopy quote. Shared Sat 2021 list (€100k/1U + €30–60k/yr) is STALE. Ask them for a written quote when — and only when — a data Service Level Agreement (SLA) is the bottleneck. ASK for a 2026 written quote. Do not buy FRAME (EnduroSat satellite bus) Max. Published unit prices depend on order size: USD9.5m (1), USD8.1m (10), USD6.7m (100), USD4.2m (1,000+) [S49]. The low end is bulk, not a single-bus or mission quote.

Dedicated Canopy-owned constellation: not viable in seed or the 18-month BG pilot. Full argument in the viability memo. Do not walk it back.

Business model

Unit of sale: a completed job — spray, survey, or duty — on a mapped plot.

Who pays: the landowner, or the duty authority.

Who gets paid:

  • Operator — majority of the money when they flew it.
  • Chem supplier — if a mix was fulfilled on that job.
  • Canopy — the take. If canopy_fleet later flies, Canopy keeps more because it owns (or leases) the steel. Split follows the flyer. Take rate to be set after paid pilots.

Take: software + match + live + receipt + cue. Hypothesis only. Take is earned on completed jobs. Take rate to be set after the first paid pilots.

Small jobs: batch nearby plots so the sortie pays. Landowner is still priced per hectare. The operator is paid on the pooled block. That is how 0.2 ha becomes first-class.

Data Cost of Goods Sold (COGS): Sentinel default, $0. Balkan / Planet only when a booked field needs it. Never a separate "Earth Observation (EO) subscription" P&L in v1.

Duty: same OS, second customer type. Ministry / fire / forest service. No weapons language. No interceptor conversation.

We never become, in v1: a satco, a drone OEM, a pesticide company, a defense prime. Option C may later lease or host a dock. That is steel we dispatch, not a factory we run.

Designed end-state: Option C

Tesla hybrid. Not Waymo. Not "we only match forever."

v1 ships A. Handshake crews. They bring their own birds. We sell the completed job. We do not sell the drone, the sat, or the tank mix.

Designed end-state is C. Third-party operators add their own birds (Airbnb + Uber). A later Canopy-owned automated dock / fleet (operator_id = canopy_fleet) turns on only when a real week is "no crew." Same test as the sat memo: own steel when paying hectares exist and hiring cannot fill the gap.

Schema now, steel later. Flyer is a field. Receipt writes operator-of-record. Same Request → Live → Receipt whether the flyer is a handshake crew or a Canopy dock. Do not fork the product.

12-month aim is the go/no-go proof: 10–30 paying jobs, live view, Bulgarian Food Safety Agency (BFSA)-shaped receipt. That is not a $1B valuation forecast. Unicorn-scale is the category ambition — the aerial job OS for any plot, any crew, later any dock. Do not write "we will be a unicorn in 12 months" as a forecast.

Full argument: research/11-platform-vs-robotaxi.md. Pricing and crop × treatment: research/12-ops-pricing-crops.md. Product note: product/option-c-hybrid-fleet.md. How existing crews’ path and receipt land on the job: product/operator-fleet-ingest.md.

Operator ingest

Handshake crews already own the birds. We do not take the sticks. We do not become an OEM. Option C: operators bring DJI Agras / XAG now; canopy_fleet dock only when a week has no crew.

Designed solution: product/operator-fleet-ingest.md. Three layers:

  1. Job contract — ours, OEM-agnostic. One job object: GeoJSON plot, weather window, customer-supplied mix, operator-of-record, receipt fields shaped like a BFSA Unmanned Aircraft System (UAS) treatment log, plus European Union Aviation Safety Agency (EASA) Specific-category fields the day we drop product. Flown path is a GeoJSON LineString with a source tag (operator_beacon | oem_kml | flighthub_mqtt | agrosol_geojson | xag_export). Receipt prefers an OEM as-applied log when present.
  2. Operator Fly app — v1 this year (BG handshake). Canopy pushes the job to the iPad. The crew flies in the native OEM Ground Control Station (GCS) they already trust. Live watch is an iPad / phone crew/device beacon — good enough for “watch the crew,” not a stick and not the aircraft path. After landing, the operator imports the OEM export and confirms hectares, dose, times. That writes a candidate receipt. No remote-control of their aircraft. The export is not automatically a legally accepted application record.
  3. OEM adapters — only when a public Application Programming Interface (API) actually exists. One adapter per vendor. Never a universal drone API. As of 16–17 Aug 2026: no public SmartFarm API (NOT FOUND; v1 = Keyhole Markup Language (KML) + beacon). FlightHub 2 OpenAPI is Dock / Matrice only — do not say it flies Agras; do not build it until we have a Dock or a Matrice survey job. XAG Open Platform is account-gated (ASK for a developer account; endpoints remain unconfirmed pending developer access). Hylio AgroSol is file GeoJSON. Pyka Pelican 2 public API NOT FOUND — not a BG v1 bird. Guardian Agriculture closed late August 2025; do not plan an adapter.

Support maps, not the flight: Sentinel via CDSE (default, $0); Planet Tasking on booked hectares; EnduroSat SpaceOps — ASK for a 2026 written quote; buying an EnduroSat FRAME Max satellite is not part of the Canopy plan; OpenWeather already in the plan.

Named suppliers we do not integrate as drone adapters (link only): Rantizo (United States (US) Drone-as-a-Service (DaaS)); DroneDeploy; Taranis; Sentera FieldAgent; Papple Aviation; AdvancedUAV / AUAV; CropCopters; American Spray Drone Coalition (ASDC); University of Nebraska–Lincoln (UNL) 2026 custom rates; Plan INFOCA (Andalusian wildfire service); European Maritime Safety Agency (EMSA); US Department of the Interior (DOI).

v1 punch: schema fields aircraft_oem, aircraft_model, gcs_app, path_source, path_geojson, as_applied_uri, live_source. First two BG handshake crews: record OEM + model; ship iPad Fly + KML / GeoJSON import; live = beacon. Do not raise to buy birds. Take % and seed size stay unset — HYPOTHESIS / ask only.

Artificial Intelligence (AI) / automation

AI is core, not a slide.

  • Cue — Sentinel / weather flags the plot and holds the window.
  • Match + batch — nearby small jobs become one sortie so the operator still gets paid.
  • Live → receipt — labelled source evidence populates BFSA field names. A crew-device beacon is not aircraft telemetry. Nobody types the log; a reviewer still accepts it.
  • Weather-window hold — software holds; Pilot in Command (PIC) can also hold.
  • Later dock dispatch — XAG XA1+LM1 / DJI Dock as vendor kit we dispatch. We do not build the Cybercab. The Dock path is FlightHub, not Agras — see the ingest memo.

Internal process: Velocity already ships with AI coding tools. The OS is built that way. The field product is the same idea: software does the match, the hold, and the artefact.

Do not value Canopy as an "AI company" with no jobs flown (research/05-valuation.md).

Capital

Spend money on the next proof. Then the next. Not on a five-year sat model. Not on a dock buy before a real week is "no crew."

Every figure that is not a published vendor number is HYPOTHESIS or "ask EnduroSat / Stripe / operators". Seed-round size is not stated in this document. There is no 12-month valuation forecast.

Now — software + first flights

  • Build Request / Fly / Live / Receipt on the stack Velocity already has (Vercel, Stripe, whatever is there). Flyer is a field. Receipt writes operator-of-record.
  • 1–2 Bulgarian operators on a handshake. No exclusivity. No fleet buy.
  • Sentinel + OpenWeather. Zero sat capex.
  • Receipt shaped like a BFSA UAS treatment log. Fields must be accepted; do not promise automatic legal compliance.
  • EASA Specific path the day we drop product.
  • Done when: 10–30 paying jobs with a live view and a receipt a reviewer would accept.

That is the 12-month go/no-go. It is already in the thesis. It is a proof, not a unicorn date.

Next — density

  • More operators. One chem partner (fulfilment, not a formulary).
  • Planet / Balkan area under management (AUM) on booked ha only.
  • Capital is working capital + two engineers + one BG ops person.
  • Not a cleanroom. Not a teleport. Not Dobroslavtsi. Not a dock purchase.

Only if the data SLA is the bottleneck

  • Then price EnduroSat Space Service or a Shared Sat slot. They fly. We buy data.
  • Get a written 2026 quote. The 2021 Portable Document Format (PDF) is not a bid.
  • Still no FRAME Max (published order-size unit prices USD9.5m–4.2m — wrong class). Still no 20-sat private fleet.
  • A small 16U cluster only when paying-ha × years > Total Cost of Ownership (TCO).

Do not raise to buy

FRAME Max. Dobroslavtsi plant. Own teleport. Interceptor work. Drone factory. A rack of docks in month one.

One line: seed money builds the OS and buys the first hundred jobs' worth of residual pixels. Satellites are a later use of profits, or a priced Space Service. Docks are a later lease when the week is "no crew." They are not the raise.

Punch list

This week

  • Call EnduroSat sales. Ask for Balkan tasking latency over Bulgaria, a data/API path, the Space Service one-pager, and current Shared Sat terms. Do not ask for a bus quote.
  • Pick one BG operator who already flies Agras or XAG. Handshake. No exclusivity. Record OEM + model on the operator profile.
  • Ship the thinnest possible Request → Live map → PDF receipt on one test plot. Live = iPad beacon. Path ingest = KML / GeoJSON import — product/operator-fleet-ingest.md.

30 days

  • Fly 5 unpaid or cheap test jobs. Write down ferry time, whether nearby plots could have batched, and how often Sentinel was cloud-killed.
  • Open Copernicus CDSE. Add Sentinel-1 for duty / flood. Do not sign Planet until a paying field needs it.
  • Put BFSA log fields into the receipt schema. Flyer is a field. Operator-of-record is a field.

90 days

  • First paid job. Stripe split-payout stub — operator gets paid, we take the rest. Take % still unset until 10 paid flights.
  • One chem distributor conversation. Fulfilment. Not a formulary.
  • One fire or forest service demo. Duty request. No weapons language.
  • Look at the week: is the gap "no crew" or "no pixel"? No crew → more operators. No pixel → then and only then price Balkan AUM.

18 months

  • 10–30 paying jobs. Live view. BFSA-shaped receipt.
  • Go / no-go on a Space Service slot — only if Sentinel + Balkan-buy cannot fill a real SLA gap.
  • Still no owned constellation.
  • Still no owned dock unless a real week was "no crew."

How we talk in the room (EnduroSat)

  1. We want residual Balkan pixels and a tasking window over booked Bulgarian fields.
  2. We want latency numbers and an API. We do not want a constellation program.
  3. Space Service / Shared Sat is a later conversation, after paying hectares exist.
  4. No FRAME Max PO. No teleport. No interceptors.
  5. If the data SLA later fails, we come back for a written 16U / Space Service quote. Not before.

Go / no-go (owned fleet)

Sats. No-go until 10–30 paying jobs and a data-SLA gap Sentinel + Balkan-buy cannot fill. Then price the service. Still don't buy the bus.

Docks. No-go until 10–30 paying jobs and a week where the gap was "no crew," not "no software." Then lease or host one dock. Still don't build a drone. Still no FRAME Max.

Sources

Same set as research/10-sat-fleet-viability.md. EnduroSat Space Service and Balkan pages accessed 16 Aug 2026. Planet 2026 published figures from observationdata.com / LandViewer / SkyWatch Jun 2026. Sentinel-2 access via CDSE (free access ≠ zero processing cost). Census 2020 holdings shape [S48]; later land-area series not headlined. FRAME Max order-size list prices in research/02-endurosat.md and [S49]. Shared Sat 2021 list is stale. Tesla hybrid and dock-kit facts: research/11-platform-vs-robotaxi.md. Ops / pricing / crops: research/12-ops-pricing-crops.md. Operator ingest (OEM APIs, path tags, v1 punch): product/operator-fleet-ingest.md, sourced 16–17 Aug 2026.