A network that turns intelligence into a graded commodity.
Ranked by performance. Driven by demand.
Plug in once. Get the best-ranked intelligence for the job.
Ask a question.
Get the top-ranked answer.
Each request resolves to an Intent, is served by the top-ranked provider for it, and is verified by validators. This is the network running live.
4.7 GW
Forecast peak wind generation, Texas, tomorrow. Above the 4 GW threshold.
Intelligence is still rough.
A rough diamond is hard to value until it is graded. Intelligence has the same problem. Thousands of models, APIs, datasets and tools can serve the same job, but their performance can differ enormously.
Telegraph continuously evaluates competing providers for each Intent, making heterogeneous intelligence comparable and rankable by performance. Paid demand can then flow through those rankings, turning intelligence from scattered supply into a graded economic commodity.
The better the supply performs, the more economic opportunity it can earn. And because Evaluators themselves compete to improve how performance is measured, both the intelligence and the market around it can keep improving.
Rough diamonds
Raw intelligence
Behind it, everyone plays their part.
Suppliers compete for paid requests, evaluators set the grading standard, validators verify and finalize the resulting rankings, and anyone can propose a better standard.
Intents · Provider ranking · Routing · Paid demand
One network. A different ranking for every job.
There is no single best source of intelligence. Telegraph ranks providers for the specific job being asked.
Weather forecast
weather-forecast
- #1Forecast model0.93
- #2Sensor API0.88
- #3Satellite dataset0.81
Weather request
shows a hyperlocal forecast
Price direction
price-direction
- #1Quant model0.90
- #2Sentiment API0.86
- #3Order-book dataset0.82
Trading query
decides whether to open a position
Fake profile detection
fake-profile-detection
- #1Identity API0.94
- #2Image-forensics service0.91
- #3Behaviour dataset0.83
Dating app
removes a fake profile before a match
A provider can rank highly for one Intent and differently for another. There is no universal provider leaderboard. These three Intents are current examples; the protocol can support many more.
Even the grading standard competes.
Providers compete on performance. Evaluators compete over how that performance should be measured. If someone builds a better evaluation method, it can replace the current standard.
Ranked by performance. Driven by demand.
Providers compete to improve the intelligence available. Evaluators compete to improve how it is measured. Validators verify and finalize the network state. Consumers create paid demand, demand rewards performance, and the opportunity attracts better participation.
As paid demand gives graded intelligence economic value, MACHINA becomes the unit through which that intelligence commodity is priced and the network around it is rewarded.
Alexandria is how humans and machines
query the network.
Telegraph is the network underneath. Alexandria is the interface on top.
What do you need answered today?
Will wind generation exceed 4GW in Texas tomorrow?
4.7 GW
Forecast peak wind generation, Texas, tomorrow. Above the 4 GW threshold.
Illustrative data.
MACHINA is the economic unit of intelligence.
Paid intelligence is priced in USDC, while Miner settlement value is converted into MACHINA - linking intelligence consumption directly to MACHINA demand.
price a company
price a project
prices graded intelligence
↻ repeat
Nobody can buy the top spot.
A ranking only means something if nobody can pay their way to the top. A single company ranking intelligence is a company selling placement, so no single company runs the scoring.
Independent Validators
verify protocol execution and finalized rankings
Competitive Evaluation
the grading standard can be challenged
Open Supply
providers compete under the same Intent-specific measurement
Real Demand
providers earn from fulfilled paid usage
Get started.
Under the hood.
For anyone who wants to go deeper. The full detail is in the whitepaper.
01WASM Evaluators
Evaluators run deterministically in a sandbox, so every validator computes the same score and no operator can alter the logic. One canonical Evaluator per Intent, replaceable at any time.
02Authenticated provenance
Every provider registers with one config file that maps its API response into a fixed, machine-readable format a contract can decode. zkTLS binds each provider to the answer it gave.
03Validator consensus
64 independent validators. Stake-weighted median, commit-reveal, BLS-aggregated, 43 of 64 to finalise.
04Ranking computation
Each Intent has its own ranking, computed continuously in the background, so a request never waits for it. Challengers are ranked on public workloads, then must pass validators running their code on hidden material.
05Spot checks
Fired from the block hash roughly every 20 seconds. A provider whose score drops 20% loses routing instantly.
06Settlement
USDC settles on Ethereum per request. 2% protocol fee, with 98% settled to providers in Machina.
07Protocol state
43 of 64 validators finalise the network state. Receipts live on Telegraph's own chain at zero marginal cost.
