Confirmation
The normal, optimistic path. An executor computes offchain and posts the result and required verification artifact onchain. A challenged portion is replayed and adjudicated onchain under the committed rules.
Intelligence,
made verifiable.
Baranos lets software verify the computation
behind an AI decision.
The problem
AI decisions need
an accountable process.
Applications use AI outputs to release funds, resolve markets and direct agents. They need to know how those outputs were produced.
Centralized LLMs and private APIs may leave the model version, evidence and execution settings hidden. A provider, or an attacker who gains access, can change those details without the application knowing.
Baranos addresses this through decentralized verification. Each job fixes the model, evidence and execution rules so independent participants can reproduce the computation and challenge the result.
The protocol
How verification works.
Confirmation mode runs inference offchain, with onchain verification and replay of disputed steps.
On Fogo
Define the job.
Commit the model, evidence, runtime and execution policy before the job runs. In Confirmation mode, a Merkle root commits to the weights while the weights remain offchain.
The same committed job has one canonical result.
Open modelWeight Merkle rootInputsCommitted evidencePolicyFixed execution rules
Offchain inferenceConfirmation mode
Result + verification artifact posted onchain
No valid challenge under job policy
If challengedReplay and adjudicate onchain
Canonical resultSettled on Fogo
Verification establishes correct execution of the agreed job. A model can still reach a wrong conclusion.
Two execution modes
Every component can execute onchain, even when the normal path runs inference offchain.
Replay
The computation executes onchain from the outset. This is the heavier mode. Baranos uses commitments, challenges and replay to enforce correct execution, without requiring a zero-knowledge proof for every inference.
Model requirements
Large models,
verifiable execution.
Baranos uses open-weight models so independent executors can reproduce a job. The initial focus is practical analysis, including wallet behaviour and evidence assessment.
Confirmation mode records a Merkle root of the weights onchain while the weights remain offchain. The protocol can replay disputed computation against those committed weights onchain.
27B
Parameters in the starting model
10T
Estimated upper verification capacity
Engineering estimate, not a demonstrated benchmark.Applications
Where applications
can use Baranos.
Applications choose the evidence and criteria for each task.
Wallet intelligence
Assess a wallet’s history of selling a token using a committed set of transaction evidence and explicit classification rules.
Attention & sentiment
Identify which people or projects are attracting attention in crypto discussions, with a fixed evidence snapshot and agreed criteria.
Prediction markets
Let traders take positions on events directly, including questions that require judgment. Participants agree on the evidence policy and interpretation criteria; disputed computation is replayed onchain against those terms.
Autonomous applications
Contracts and agents can use a settled model result to decide whether evidence meets the conditions for an action, such as authorising a payment.
Accuracy depends on the chosen model, evidence and policy.
The settlement layer
Settlement on Fogo.
Results across chains.
Fogo's fast blocks and low transaction fees help make frequent verification and disputed-step replay practical. Baranos targets minute-scale results and dispute resolution, with timing dependent on the workload and challenge policy.
Baranos currently runs on Fogo testnet. Settled results become onchain state that applications can read.
Through a Wormhole messaging integration, contracts on other supported chains could receive and use results settled by Baranos.
Baranos settles on Fogo. Wormhole carries an authenticated message; the destination contract verifies it before using the result.
Cross-chain delivery has its own latency and security assumptions.
Resources
Read more
about Baranos.
The product introduction, technical explainer, one-page overview, deck and whitepaper.
Explore the resources