Get Started
Testnet Overview
The Lightchain AI Testnet is the public network for building and testing on Lightchain AI before mainnet. It is built from the same Lightchain Geth and Prysm forks as mainnet, and it adds a sharded execution layer that is live only on the testnet.
The testnet is in public beta. It is monitored by the Lightchain team, and parts of it are still changing, including the sharded execution layer described below. Expect occasional resets, endpoint changes and short interruptions. If you run into a problem, please report it: every report helps.
Testnet LCAI has no real-world value. Get it for free from the faucet.
The testnet was reset from a new genesis on 2026-10-28 14:46:36 UTC. The chain ID is unchanged (8200), but contracts, balances and transactions from before that date no longer exist. Keep your deployment scripts version-controlled so you can redeploy.
Network Details
The public endpoints publish no fixed rate limit and may be rate-limited during periods of high demand. If you are throttled, consider running your own testnet node.
Chain Parameters
The Basis column says where each value comes from: Protocol values are set in the chain configuration, Observed values are read from the live testnet.
During the beta, all 21 testnet validators are operated by the Lightchain team. The testnet is not yet decentralized.
Architecture
The testnet has two layers:- The Lightchain L1 (chain
8200). A proof-of-stake chain built from Lightchain forks of Geth (execution) and Prysm (consensus). It runs smart contracts, the AI inference protocol and staking, and it is where finality comes from. Your wallet connects here. - The sharded execution layer (testnet only). Seven additional EVM chains ("shards", chain IDs
82001–82007). A replicated sequencer orders their transactions and anchors every shard block to the L1. Shards add execution capacity in parallel, without making every L1 validator execute every transaction.
CodeTEXT
How a shard transaction is processed
- Sequencing. A shard transaction is signed with that shard's chain ID, like a transaction on any EVM network. The sequencer checks the signature, the nonce, a conservative lower bound on the sender's balance, and the minimum fee (1 gwei). It then writes the transaction to its replicated log and returns the transaction hash.
- Ticks. Every 500 ms the sequencer closes a tick: an ordered, signed batch of transactions for each shard.
- Execution. Each shard's driver fetches its batch, checks the signature, and has the execution client build exactly that block. The driver has no discretion: every shard block is a deterministic function of the signed batch. A transaction that fails at execution (for example because its sender's balance was spent in the meantime) is recorded as dropped, and its receipt returns an error so the sender can resubmit.
- Anchoring. About every 2 seconds (once per L1 block), the sequencer writes the new shard blocks' state, exit and batch roots to the ShardRegistry contract on the L1.
- Bridging. LCAI moves onto a shard through ShardBridge on the L1. A deposit is credited on the shard after the L1 deposit is final. LCAI moves back through an exit on the shard, which is claimed on the L1 against the anchored exit root.
Components
Shard endpoints
You can check any shard block against what the L1 recorded for it. The shard dashboard shows the exact commands under Verify it yourself.Other L1 components
- AI inference protocol. AI inference runs through protocol contracts (
SessionManager,JobRegistry,WorkerRegistry,AIConfig) and independent workers. See AI Inference Architecture and the Developer API. - Blobs. EIP-4844 blobs with KZG commitments carry encrypted inference payloads, with a separate blob fee market.
What is not live yet
We publish the limits of the beta so you can judge what the testnet does and doesn't show:Performance
Every throughput number on this page was measured on the live testnet, not on a simulator or a separate development network. Measured figures are counted from the blocks the shards actually produced, not from what the load generator sent.
What "TPS" means here. Throughput depends on how much gas each transaction uses. The figures below are for plain LCAI transfers (21,000 gas each) unless a row says otherwise. Heavier contract calls fit fewer per block.
At a glance
Results by stage
The load went up in steps of 5 minutes each. Every shard produced 2 blocks per second at every step, including when blocks were completely full. Why throughput above about 14,000 tx/s doesn't last. Shards use the EIP-1559 fee market: when blocks are more than half full, the base fee goes up with every block. Up to 14,000 tx/s the blocks stayed at about half full and the base fee didn't change. At 18,000 tx/s the base fee rose for the whole stage, and at 22,000 tx/s it rose fast enough that the test wallets could no longer pay after 2 minutes, which ended the test. This is the fee market working as designed: it prices out load above the target. So about 14,000 tx/s is the rate the shards can sustain at stable fees, and about 21,000 tx/s is a short peak. A planned 30-minute stage was not run, because the test wallets had run out of LCAI. The sustained figure above is therefore over 5 minutes.Method
- Network: the live testnet: the 7 shards (
82001–82007), the 3-node sequencer, and the L1 (8200) they anchor to. Nothing was reconfigured for the test. - Load: Lightchain's
stressload generator, sending signed transactions from 3,000 funded test wallets per shard at a fixed offered rate, raised in steps of 5 minutes. It ran on the sequencer's host and sent to the sequencer over the host network, so the results measure the chain, not the public internet. - Counting: included transactions per second come from the shard blocks themselves (transaction count and gas used per block). The first 30 seconds and the drain at the end of each stage are excluded.
- Health: throughout the test, every validator machine was checked every 30 seconds (finality, peers, validator errors, attestations), together with the sequencer's anchoring lag, errors and disk. The test would have stopped automatically if finality fell more than 4 epochs behind.
Check the numbers yourself
Shard blocks are public, so anyone can count the transactions in them. For example, the latest block of shard 1 (replace"latest" with a hex block number such as "0x9c777" for a specific block):
CodeBASH
Limits of these numbers
- Shard load came from a single source with transfers only. Real traffic mixes transaction types and arrives through public RPC endpoints, which have their own limits.
- During the beta, the sequencer and all 7 shards run on one machine (an AMD EPYC 9254 server). The numbers are a measurement of that setup, not a ceiling of the design.
- Latency is measured by the load generator, from sending a transaction to seeing it in a shard block. It doesn't include L1 anchoring or finality.
- The L1 is not yet at full health outside the test: in the hours around it, about 0.7% of L1 slots had no block. We are investigating. So far the missed slots haven't affected finality or shard block production.
- The shards are not yet open to public transactions (see What is not live yet), so these figures describe capacity, not current usage.
Getting Started
Follow these guides in order to get up and running on the testnet:- Connect to Testnet — Add the Lightchain AI Testnet to your wallet.
- Get Testnet Tokens — Claim free LCAI from the faucet.
- Deploy a Smart Contract — Deploy your first contract on the testnet.
- Interact with Contracts — Read and write to your deployed contracts.
- Verify a Contract — Verify your contract source code on the block explorer.
Reporting Issues
The testnet is in beta and is monitored by the Lightchain team. If something doesn't work as you expect, please tell us in Discord — #dev-support. A good report includes:- what you did, what you expected, and what happened instead;
- the time (UTC) and the endpoint you used (for example
https://rpc.testnet.lightchain.ai); - the transaction hash, block number or contract address, if there is one;
- the exact error message, or a screenshot.
Security issues: don't post vulnerabilities in public channels. Contact the team privately through Discord first.
Testnet Changelog
Last updated: 1 November 2026