NEAR Protocol

NEAR's current history page traces the partnership to 2017, when Illia Polosukhin and Alexander Skidanov formed NEAR.ai around a concrete ambition: models that could translate natural-language instructions into code. Polosukhin brought machine-learning research, including co-authorship of the Transformer paper; Skidanov brought distributed-database engineering. This is an official retrospective and should be read as the project's account of its own origin, rather than independent proof that every present AI claim was planned at the time.

According to that account, the models and available GPU capacity were too early, while paying a global group of student contributors revealed how awkward cross-border settlement could be. The pair turned toward blockchain and in 2018 founded NEAR Protocol. The causal thread was therefore practical rather than mystical: an AI product ran into coordination and payment constraints, and its builders decided to create infrastructure they believed developers could actually use.

NEAR's persistent mainnet state began on 22 April 2020, but the first phase was Proof of Authority, not an unrestricted public network. The NEAR Foundation operated the initial nodes, controlled most transfers and created accounts for approved developers. The later restricted phase moved validation toward an external validator set while ordinary transfers remained constrained. These details matter because a bare '2020 mainnet launch' hides who could act and which functions were live.

The unlock was itself a governance event with a narrow purpose. Validators and delegators signaled that transfers could open, and unrestricted operation followed in October 2020. That vote did not establish a permanent direct-democracy model for every token holder. It demonstrated the network's stake-weighted path out of launch restrictions, while later protocol upgrades continued to depend on validator adoption of nearcore releases.

The launch network used one shard even though the architecture had been designed around sharding. Simple Nightshade changed that in November 2021 by splitting mainnet into four shards while keeping the developer experience largely unified. It was Phase 0: more parallel capacity and a foundation for later changes, rather than the final Nightshade design described in early papers.

Nightshade 2.0 reached mainnet in August 2024 with stateless validation. Chunk validators could verify work from state witnesses without keeping every shard's full state, a prerequisite for scaling the validator workload across more shards. NEAR One's accompanying roadmap was unusually explicit that the first release did not immediately improve performance. Later nearcore releases stabilized dynamic resharding; a live final-state query on 4 September 2026 reported protocol version 86 and ten shard IDs. The implemented design can split a shard after configured thresholds, but the cited proposal does not establish automatic merging or unbounded throughput.

By 2024, NEAR's chain-abstraction thesis had produced two concrete mechanisms. Chain Signatures lets a NEAR account or contract request a jointly generated MPC signature for a derived account on another network. The target-chain transaction still has to be constructed and broadcast, and the mechanism is one-way signing rather than a general reader of external state. It reduces key fragmentation without making cross-chain execution trustless by definition.

NEAR Intents shifts the interface from transactions to desired outcomes. A person or AI agent signs a request, off-chain solvers compete to quote execution, and an accepted result is verified and settled through a contract on NEAR. The swap path and developer interfaces are real deployments. NEAR's larger claim—that private, user-owned agents will coordinate money, credentials, compute and arbitrary services through this layer—is a strategic program whose breadth still depends on products, liquidity, solver diversity and users that do not yet come from the base protocol alone.

At protocol level, NEAR pays transaction and computation costs, backs storage accounting, transfers value and secures validators through staking. Current official validator documentation targets annual issuance equal to 2.5% of total supply. Fees remaining after the contract rebate are burned, so net growth depends on network use, but holders should not turn that mechanism into a guaranteed-deflation or guaranteed-yield story. Rewards compensate validation; they are not corporate dividends.

Decision power is more fragmented than the phrase 'governance token' suggests. Validators vote by stake on protocol-version upgrades. House of Stake is a separate system in which users lock NEAR for veNEAR and vote within a proposal and treasury framework that also includes reviewers, configurable contract owners and council vetoes. The NEAR Foundation remains a separate organization with strategy and resources. Ordinary NEAR ownership therefore does not automatically confer a direct ballot over every code change, House of Stake disbursement or Foundation decision, and the current identities behind every administrative role still require contract-by-contract verification.