A drone brain trained by everyone, owned by no one.
Flock is an open flight-data network. Pilots submit signed flight logs, telemetry and edge-case footage; the network verifies each submission and mints FLK to the pilot on a fixed, decaying schedule. The pooled data trains a line of open-source flight models anyone can download. Reward weights, treasury and upgrades are decided by token vote — a DAO in which no one holds a special seat. A flock has no leader. That's by design.
- 01Fly
- 02Submit
- 03Verify
- 04Earn
- 05Train
- 06Govern
Epoch
14 · month 2 of 1Rewards settle hourly, epochs monthlyVerified submissions today
—Across all pilotsFLK minted this epoch
—Of a 3.41M pilot poolActive pilots
2,847Submitted in the last 30 daysCurrent open model
FM-2.3Apache-2.0 · weights on the ledgerHow it works
Data in. Tokens out. Models for everyone.
The loop is the product. Every verified flight makes the dataset bigger and stranger; every retrain makes the open models better; better models pull in commercial users, who burn FLK to license training runs; the burn gives the token a job beyond speculation, which pays the next pilot to fly.
Fly
Any pilot, any aircraft. A signing key on the drone or the companion app seals each log at capture.
Submit
Flight logs, raw telemetry, imagery hashes, incident reports. Push from the field or batch from the desk.
Verify
Staked validators dedupe, cross-check against weather and ADS-B, and score quality 0–1.
Earn
Your share of the epoch's emission equals your share of the network's verified reward points.
Train
The pool retrains the Flock Models on a public manifest. New weights ship under Apache-2.0.
Burn
Commercial users burn FLK for training-run credits. Hobbyists use the models free, forever.
Prior art
What Helium proved, and what it warned.
In 2019, Helium paid ordinary people in HNT to put LoRaWAN hotspots in their windows. It worked: strangers built a global wireless network no company could have deployed, and early adopters were rewarded most because emissions were largest at the start. Flock borrows that spine — permissionless hardware at the edge, tokens minted for verified physical work, a schedule that favors the people who show up first.
It also borrows the caution. Helium's coverage grew years ahead of its usage, hotspot spoofing had to be fought with challenge games, and token price did more recruiting than network demand ever did. So Flock ties minting to verified data rather than mere presence, builds the challenge system in from day one, and gives the token a native sink — burn-for-training — so demand is measured in model runs, not sentiment.
Why no one owns it
The sky shouldn't have a landlord.
Low-altitude airspace is about to become an economy — delivery routes, rooftop pads, charging corridors, the data that makes it all safe. If a government or a single company ends up owning that layer, everyone else becomes a tenant of their own sky. Flock's answer is structural: the network is owned by whoever does the work, and there's work for everyone.
The argument
- Not governments. Regulators should set the safety rules of the air — and sit on the ledger as validators — without picking which companies get to profit from it.
- Not companies. Operators fly in the airspace; they shouldn't own the commons underneath — the data, the models, the economic layer. Air Pixels included.
- Everyone underneath it. Deliveries land on your street, drones cross your roof, your weather hits their logs. If the drone economy runs over your head, you deserve a way to hold a piece of it.
Literal airdrops
Some deliveries carry a FLK claim code — an airdrop in the oldest sense. Scan the box, claim your stake, and the person who ordered socks walks away holding a sliver of the network that delivered them.
Ways in
Today's priority
Pilots
Fly, submit, mint. The emission curve exists to get the dataset built, so the biggest rewards go to the people flying right now.
Property
Rooftop hosts
Real estate agents, developers, homeowners: list pad space, charging perches or a weather sensor on the network and earn FLK every time a flight uses it.
No drone required
Neighbors
Order a delivery, confirm a landing zone, report an obstruction, catch an airdrop. Attention is work the network needs — and pays for.
Industry
Drone companies
Burn FLK for training runs, stake validators, fund grants. Same door as everyone else — capital buys usage, never control.
Infrastructure
Validators
Stake FLK, run a node, score and challenge submissions. The work earns the 12% slice; rubber-stamping junk gets the stake slashed.
Open source
Builders & researchers
Log parsers, field apps, retrains, eval benchmarks — treasury grants fund the tooling, and every improvement is traceable to its author.
Human in the loop
Annotators
Label landing zones, confirm near-miss footage, review flagged imagery from a phone. Verified labels mint exactly like flights do.
Hardware
Manufacturers
Ship signing keys in firmware so every flight is submission-ready out of the box — and earn a sliver each time your airframes contribute verified data.
Institutions
Insurers & regulators
Run observer validators, underwrite and audit from the same ledger everyone else reads. Commercial use is paid for in burns, like anyone's.
Earning
The network pays for what it can't simulate.
Reward weights express what the models are missing. A thousand sunny waypoint missions teach less than one autorotation in sleet — so the schedule pays accordingly, and the DAO retunes the weights as gaps close.
| Submission | Weight | Why it's worth that |
|---|---|---|
| Flight logRoute, timestamps, outcomes, signed at capture | 1.0 pts | The baseline unit. Establishes provenance and anchors everything else you attach to the flight. |
| Full telemetryIMU, GNSS, barometer, motor currents at 10 Hz | 2.0 pts | Teaches control models how real airframes behave between the waypoints — gusts, vibration, battery sag. |
| Consented imageryGeotagged stills or video, hashed on-device | 3.0 pts | Perception data for landing-zone and obstacle models. Faces and plates are redacted before the hash. |
| Edge-case flagsNight, snow, >20 kt wind, urban canyon, GNSS-denied | +4.0 pts | The long tail every simulator misses. One flagged flight in freezing fog outweighs a week of fair-weather laps. |
| Incident reportNear-miss, failsafe trigger, forced landing, flyaway | +8.0 pts | The rarest and most valuable class. Paying for honesty about failures is how the safety models get good. |
Novelty multiplier
×0.7 – ×1.8Set by coverage density, like Helium's hex scaling: data from a saturated test field earns less, first flights over sparse terrain earn most.
Quality score
×0.0 – ×1.0Validators score completeness, sensor sanity and corroboration. Junk scores zero and mints nothing — volume alone is worthless.
Genesis bootstrap
×2 · first 90 daysA one-time multiplier while the network is empty, when data is most needed and the token is worth the least. After that, the curve does the rewarding.
Flights per week
What you submit (logs always included)
Where you fly
When you join
Estimated earnings
— FLK / epoch
- Your reward points / epoch—
- Network points / epoch—
- Your share of emission—
- Pilot pool this epoch—
Concept math, not a promise. Assumes the modeled adoption curve (2,500 pilots in year 1, compounding ~2.2× yearly) and average activity from everyone else. No dollar figures on purpose: FLK has no price, and this page isn't here to invent one.
The token
The schedule is the contract.
250 million FLK, ever. Emissions unlock automatically and decay 20% a year — largest when the network most needs convincing, smallest once it can stand on demand. No premine, no founder tranche, no admin key that can print more. Early pilots are paid by the curve, not by promises.
Emission & circulating supply, years 1–15
Where each epoch's emission goes
- Pilot rewards verified data
of emission80% - Validators & model builders staked work
of emission12% - DAO treasury grants by vote
of emission8%
If you join in year 1
Already minted
0%of max supply, before your first flightStill ahead
100%mintable after you arriveThat year's emission
50MFLK, shared by that year's pilotsMint side
Pilots mint by flying.
Emissions only enter circulation against verified submissions. No flight, no mint — the schedule caps the rate, the work claims it.
Burn side
Companies burn to train.
Commercial training runs, fine-tunes and dataset exports cost Data Credits, bought only by burning FLK. Usage shrinks supply; the models stay free for individuals.
Verification
Paid data invites fake data. Plan for it.
Every network that has paid for physical data has been paid in fakes. Helium answered hotspot spoofing with challenge games; Hivemapper caught GPS-spoofed dashcam runs with sensor cross-checks; DIMO starts from hardware-signed vehicle telemetry. Flock expects GPS replays and simulator dumps on day one — every submission runs this gauntlet before a single token mints.
Signed at the source
A per-aircraft key seals each record as it's captured, Proof of Flight style. A log assembled after the fact carries the wrong signatures.
Deduplicated
Record hashes are checked against the whole pool. Resubmitting a flight — yours or anyone's — collides and mints nothing.
Corroborated
Claimed wind against weather archives, claimed position against ADS-B and Remote ID, claimed dynamics against the physics of the declared airframe.
Challenged
Validators issue random challenges: re-derive a segment, produce a sibling hash, prove the sensor noise floor. Simulator output is too clean to pass.
Staked
Validators post FLK to score submissions and lose it for approving fakes a challenge later exposes. Honesty is the profitable strategy.
Anchored
Accepted batches are sealed into Merkle roots on the shared ledger, so the dataset's history can't be quietly rewritten — by anyone.
The models
The output is open source. All of it.
The Flock Models are the network's reason to exist: flight intelligence trained on the pool and released under Apache-2.0. Weights, training manifests and eval results are hashed to the ledger, so every release is reproducible and every improvement is traceable to the data — and the pilots — that caused it.
| Release | Trained on | Landing-zone assessment | Wind-field estimation | Failure prediction |
|---|---|---|---|---|
| FM-0.1Genesis · simulated + seed data | 12k flights | |||
| FM-1.0First full retrain · month 6 | 184k flights | |||
| FM-2.0Edge-case weighting lands · year 2 | 1.1M flights | |||
| FM-2.3Current · incident corpus matures | 2.9M flights |
Figures are illustrative — the point is the shape: the biggest jumps follow the data the reward schedule chased. FM-2's edge-case weighting and FM-2.3's incident corpus exist because the network paid 4 and 8 points for exactly that flying.
Governance
Nobody's hand on the tiller. Including ours.
One FLK, one vote. Parameters pass at 66% with 10% quorum; touching the emission curve itself takes 80%, on purpose. Proposals, votes and treasury flows are on the same ledger as everything else.
The DAO decides
Everything tunable.
- Reward weights and multipliers, as model gaps open and close
- Treasury grants: parsers, tooling, compute for retrains
- Validator admission, stake sizes and slashing terms
- Model release cadence and eval benchmarks
- Protocol upgrades, including replacing this very design
The DAO cannot
Everything else.
- Mint beyond the schedule — 80% supermajority or it doesn't move
- Revoke the Apache-2.0 license on shipped weights, ever
- Deanonymize pilots or unredact imagery already pooled
- Seize or claw back tokens already minted for verified work
Air Pixels' role
Pioneer, then just another pilot.
- Author the v0.1 protocol and seed the genesis dataset — then submit it to the first vote like any other proposal.
- Run one validator among many, under the same stake and slashing as everyone else.
- Earn FLK only the way pilots do: by flying and submitting. No premine, no veto, no admin key.
- If the DAO votes Air Pixels out of the validator set someday, that's the system working.
Genesis registry
Register interest without becoming a customer.
A network owned by no one shouldn't start life as somebody's mailing list. Register with a wallet and the registry holds a signature — cryptographic proof a person wants in — with no name, no email, no profile for anyone to own or sell. Prefer email? It's stored as a one-way hash commitment.
What registering gets you
- A place in line. Genesis-registered identities are first eligible for the 90-day ×2 bootstrap window — pilots, hosts and neighbors alike.
- A voice before launch. Pre-genesis FIPs — reward weights, validator admission, the legal wrapper — go to registered identities for straw votes.
- Nothing else. No tokens, no allocation, no promise of value. FLK doesn't exist, and registering doesn't change that.
Registry entryEMPTY
- Identity
- —
- Method
- —
- Roles
- —
- Registered
- —
- Position
- —
Commitment · awaiting signature or email hash
How will you join the network?
Choose an identity
Your key signs one message in your wallet. The registry keeps the signature, never the key — pseudonymous by construction. No wallet? We'll mint a throwaway key right here in your browser.
Stored as a SHA-256 commitment — at launch, proving you hold the email unlocks the registration. A plain copy would be held encrypted only until genesis, then deleted.
Founding statement
Built to be walked away from.
Air Pixels built Flock because safety in the low-altitude sky is a data problem. The models that keep drones clear of helicopters, powerlines and playgrounds will only be as good as the flying they learn from — and no single company will ever fly enough. The sky this all happens in is a commons; the intelligence that keeps it safe should be one too.
So the framework asks for everyone's flying and pays for it. That only works if contributors trust the ledger more than they'd have to trust us. An owner — any owner, however well-meaning — can reprice the work after it's done, resell the commons, or quietly tilt the models toward its own fleet. The cure isn't a benevolent owner. It's no owner. The emission schedule is code, verification is automated, the models ship under an irrevocable open license, and every rule that can change changes by vote.
We wrote v0.1 and we'll argue for our proposals like anyone else — holding 0% of the supply, running a validator seat we have to keep winning back. If Flock works, the sky gets safer and nobody, including us, decides who it works for.
— Air Pixels · October 2026
Surviving its users
Every incentive invites an exploit. Name them now.
A network nobody controls still has to be defended — by structure, not by trust. The attacks below are the obvious ones; the honest column is what each defense doesn't fix.
| The exploit | The structural answer | What it admits |
|---|---|---|
| Buy the voteAccumulate FLK, rewrite the rules | Decaying emissions mean dominance can't be minted late, only bought in the open. Every parameter change sits in a 14-day timelock; the emission curve itself needs 80%. | Token voting still favors capital. If capture ever wins, the fork is the final veto — manifests and weights are already public. |
| Validator cartelCollude to approve junk data | One seat per entity, stakes capped at 10% of the set, challenges assigned randomly, slashing on any approval a later challenge exposes. | Cartels move faster than elections. The challenge budget is permanent for a reason. |
| Farm the rewardsFly the same safe field a thousand times | Novelty multipliers decay per covered hex, quality scoring zeroes near-duplicates, per-region emission caps stop industrial-scale laps. | Every cap breeds a cleverer farm. Retuning weights is the DAO's forever job, not a launch task. |
| Air Pixels overstaysFounder gravity becomes soft control | Our validator seat expires every 12 months unless re-elected. No premine to vote with, no admin key to refuse to hand over. | Influence outlives keys. Read our proposals as skeptically as anyone's. |
| Drain the treasuryGrant spam and friendly invoices | Grants are milestone-escrowed, capped per epoch, and clawed back by vote when milestones slip. | Committees can be charmed. Small caps protect better than good intentions. |
| Let it rotApathy, low turnout, stale weights | Delegated voting, and parameters that auto-tune inside voted bounds so safety never waits on a quorum. | Automation sustains; it doesn't inspire. A network nobody cares about deserves to be forked by people who do. |
What keeps it alive
Demand before genesis
Committed burnersLaunch waits until commercial partners have contracted to burn FLK for training runs — usage on day one, not Helium's empty-coverage years.
Income after emissions
5% burn titheAs the curve decays, a tithe on commercial burns replaces emissions as the treasury's income — retrains and challenges stay funded forever.
Succession by default
Everything forkableIrrevocable licenses, public manifests, re-electable seats. The credible threat of leaving is what keeps every participant honest, including us.
Open questions
What v0.1 doesn't answer yet.
Is FLK a security?
Tokens minted for work look different from tokens sold for money, and Helium navigated this for years — but "probably fine" isn't a legal strategy. The honest answer: jurisdiction by jurisdiction, with counsel, before a single token mints.
Whose backyard is in the dataset?
Flight data is location data. On-device redaction and hashing help, but consent for imagery over private property, GDPR-grade erasure from a pool models already trained on, and incident reports that implicate third parties all need real answers.
Can verification stay ahead of spoofing?
GPS replay, simulator dumps and sensor-noise mimicry will improve exactly as fast as the rewards justify. The challenge system is a treadmill, not a wall — budget for the arms race permanently.
Does demand actually arrive?
Helium's hardest lesson: supply-side incentives work almost too well, and demand lags. If nobody burns FLK for training runs, the token is just points. The burn side needs committed launch partners, not hope.
Who pays for the retrains?
Training on millions of flights isn't free. Treasury-funded compute grants work on paper; whether an 8% emission slice covers frontier-scale runs — and who's accountable when a release regresses — is unsettled.
Can a DAO do safety-critical work?
Token votes tune reward weights well. Whether they should gate the release of a model that real aircraft will fly with is harder — a safety council with a narrow veto may be worth the purity cost.
Roadmap
Let go of the string on schedule.
Air Pixels can carry Flock to the launch site; it can't keep holding the balloon. Each phase below removes one of our responsibilities and names the problem most likely to appear there. Phases advance on proof, not calendar dates — the gates are the map the network follows after we let go.
- PHASE 0
Draft
Air Pixels: author
The framework in public — this page, straw votes from the genesis registry, the legal wrapper. Reward weights and the emission curve get argued in the open before any of it is code.
What gets hard
Securities analysis, jurisdiction by jurisdiction. "Probably fine" doesn't mint.Gate to advance
v1.0 spec ratified by registry straw vote; counsel signs off per launch market. - PHASE 1
Genesis
Air Pixels: operator
Testnet with real flights. The reference app ships signing keys, the first pilots mint under the ×2 bootstrap, and Air Pixels runs ingestion and the seed validators because someone has to. The only phase with training wheels.
What gets hard
Format fragmentation — most consumer flight logs are locked in proprietary formats.Gate to advance
100k verified flights · 3+ independent validators live · burn contracts signed. - PHASE 2
First retrain
Air Pixels: trainer
The data pipeline earns its keep: dedupe → corroborate → score → anchor, with no human in the approval path. Pilots keep their raw data; the network stores hashes, features and consented extracts. FM-1 ships from a public manifest.
What gets hard
Storage and training compute scale faster than an 8% treasury slice.Gate to advance
An independent team reproduces FM-1 from the public manifest alone. - PHASE 3
Open doors
Air Pixels: one of many
Every way in goes live — hosts list pads, annotators label, manufacturers ship keys in firmware, institutions run observer nodes. Burn-for-credits opens, and demand gets measured instead of assumed.
What gets hard
Demand lag — Helium's lesson arrives right on schedule.Gate to advance
Burns ≥ 25% of emissions for two consecutive epochs. - PHASE 4
Hands off
Air Pixels: observer
Our ingestion is decommissioned and re-implemented by validators; parameters auto-tune inside voted bounds; our validator seat stands for its first re-election. The string gets let go here, not at a press event.
What gets hard
Founder gravity — influence that outlives the infrastructure.Gate to advance
90 consecutive days with zero Air Pixels-privileged actions on the ledger. - PHASE 5
Released
Air Pixels: pilot
A standing state, not a finish line. The DAO can retune, re-staff or replace the design itself — the map is instilled, the string is cut. Air Pixels keeps flying and submitting, one pilot among thousands.
What gets hard
Apathy. A healthy protocol still needs people who care enough to vote.Gate to advance
None. From here, the roadmap belongs to the vote.