Frequently Asked Questions

Savitri Network is the first Layer 1 blockchain where Federated Learning aggregation runs at the consensus layer — not as a smart contract on top, not on an off-chain coordinator. The difference from other "AI + blockchain" projects: · Most projects bolt smart contracts on existing chains → chain only sees a hash of the model → off-chain aggregator is still a single point of trust · Savitri's consensus sees the model weights → aggregation runs IN consensus → no off-chain coordinator to trust The mechanism: the same Proof-of-Unity score that elects block proposers also weights FL gradient updates during the federated round. One mechanism. Two jobs. In practice: testnet live with 45,725 TPS peak / 485ms finality on 25 cloud nodes. 1 enterprise customer in production for 1 year (academic / government deployment). TensorFlow Federated, PyTorch, Flower — all run on top via savitri-fl bridges.

Three problems FL has at production scale that no framework solves: 1. Data Poisoning — bad gradients corrupt the global model; detection requires trust in a reference 2. Free-Rider Problem — without quality-weighted compensation, rational actors free-ride 3. No Provenance — no auditable lineage for compliance (EU AI Act 2026, HIPAA, GDPR) Smart contracts on Ethereum/Solana can't see model weights. Storing a 50MB ResNet update on Ethereum costs ~$10,000 in gas. To do FL aggregation IN consensus, you need new L1 work — not a fork.

TensorFlow Federated, PyTorch, Flower — all via savitri-fl bridges (alpha). Custom FL implementations also supported via direct savitri-fl Rust SDK. You don't rewrite your FL workload. You add 4 API calls: register model, submit gradient, query state, claim quality-weighted rewards. The aggregation runs at consensus layer.

Every node on the network accumulates a "trust score" from 0 to 1000. The score measures: are you online when needed? Are the things you propose valid? Do you participate actively? Nodes with a high score earn more. Nodes with a low score (< 300) receive no rewards — but can get back on track without losing everything. Think of it as an Airbnb rating for your node: the more reliable you are, the more you earn.

No. There is no minimum stake requirement. To run a Lightnode you need: · A laptop or smartphone with 4GB of RAM · A 1 Mbps internet connection · The app or installer (both free) SAVI rewards arrive automatically once your node accumulates a sufficient PoU score (≥ 300/1000).

Yes — it's an architectural choice, not a marketing promise. The Lightnode client is written in Rust with Flutter, compiled natively for ARM64 (iOS and Android). It's not a WebApp in a wrapper — it's a real node. It has three battery consumption modes: Active: full participation (while charging) Balanced: default, reduced heartbeat Power Saving: heartbeat only < 20% battery Runs in the background. Doesn't occupy the screen.

No. Savitri is on testnet. The tokens you earn now are test tokens — they have no economic value and are not transferable. When mainnet goes live (after the audit remediation is complete), that will change. We do not make listing announcements before concrete conditions are in place.

Base: 100 SAVI per epoch. With maximum PoU score (1000/1000): 300 SAVI per epoch (3.0x multiplier). Current epoch: 300 seconds (5 minutes) → ~288 epochs/day Monthly estimate (score 850/1000, uptime 90%): ~250 SAVI/epoch × 0.90 × 288 × 30 ≈ ~1,944,000 SAVI/month Note: approximate calculation on testnet. Use the interactive calculator at /run-node for your specific configuration.

It depends on how long you're offline: Brief interruption (< 1 epoch): Small reduction in availability score. Rewards are reduced proportionally in that epoch. No permanent effect. Extended downtime (days): Score drops by ~10% for each prolonged offline period. You are not excluded from the network — but you earn less until you recover your uptime. The system distinguishes a normal restart from a prolonged absence. Home connection issues are expected and tolerated.

For the Lightnode: no, it's not required. The lightnode connects to the P2P network via Kademlia DHT and maintains connections even with a dynamic IP. For the Masternode: recommended. Masternodes participate in BFT consensus with direct voting — a static IP improves reachability by other masternodes and reduces latency in voting rounds. It's not mandatory, but it improves Availability and Latency scores.

3 options: 1. Installer Wizard — for those who don't want the command line Windows / Linux / macOS — download from /run-node 2. Docker — for those who prefer containers

cd docker && docker-compose up -d
# Starts 2 MN + 2 LN + Grafana in one line

3. Build from source — Rust 1.82+

cargo build --release
# For those who want to audit the code

Full guide: /run-node

Yes, with Docker you can start multiple containers on the same host. The default docker-compose starts 2 masternodes + 2 lightnodes on the same PC — it's the standard testing setup. Each node has: · Separate ed25519 keys (dedicated -keys volume) · Independent PoU score · Separate P2P ports (4001/4002 for LN, 4021/4022 for MN) Note: each node consumes RAM independently. With 4 nodes on a 16GB PC, you have ~4GB per node.

The keys are in the Docker volume (*-keys) or in the installer folder (~/.savitri/keys/ by default). If you back up the keys before losing the PC, you can restore the node on different hardware with the same identity and the same PoU score. If you don't have the backup: you lose the accumulated score and reputation. There is no recovery mechanism — it's by design (security). Seed phrase backup at installation — mandatory.

Not directly EVM-compatible in the sense of executing EVM bytecode. But our smart contract standards are designed to feel EVM-familiar to an Ethereum developer: Savitri-20 → same API as ERC-20 (mint, burn, transfer, approve...) Savitri-721 → same API as ERC-721 Savitri-1155 → same API as ERC-1155 The runtime is Rust-based, not EVM — which means: ✓ Faster ✓ No gas estimation ✗ Solidity bytecode cannot be imported directly ✗ Hardhat/Foundry tools don't connect natively If you use the Savitri SDK, the logic is familiar. The code is not 1:1 portable.

Yes. RPC base URL:

http://testnet.savitrinetwork.com:8546

(or any node running locally) You can use it right now:

curl http://testnet.savitrinetwork.com:8546/savitri/health

Free testnet tokens from the faucet:

POST /savitri/faucet/claim
# 1 claim per address per 24 hours.

docs.savitrinetwork.com — complete documentation: · Consensus spec (PoU, BFT, DAG) · RPC reference (all 13 endpoints) · Rust SDK (savitri-sdk = "0.4") · Flutter SDK (savitri_flutter: ^0.4.0) · Smart contract guide (Savitri-20/721/1155) · Node setup (Docker, installer, source) Quick reference also at /developers — with live playground.

Yes — the repository is public, MIT + Apache 2.0. Contributions welcome via pull request on GitHub. What you need: · CONTRIBUTING.md in the repo for guidelines · Open issue before a large PR · Tests for every new feature Active bug bounty: valid vulnerabilities are rewarded in SAVI. Details at /security.

100,000,000 SAVI. Fixed. Important note: some previous communications indicated "2 billion" — that figure was incorrect. The correct supply hardcoded in the protocol is 100,000,000. Verifiable in the tokens.toml file in the public repository.

No confirmed date. Prerequisites that must be met before listing: 1. Audit remediation complete (0 open critical findings) 2. Post-remediation final audit passed 3. Mainnet live and stable The first planned listing is on native DEX, then at least 1 CEX. Any listing announcement not coming from official @savitrinetwork is unauthorized.

The /team page will be published with verified team data before the mainnet launch. We do not publish partial or unverified information. For institutional investors with serious interest: [email protected] — response within 5 business days.

Yes. Completed in February 2026 by an independent third party. 68 findings (16 critical, 22 high, 18 medium, 12 low). Remediation in progress — ETA Q2 2026. Full details at /security, including the live remediation status and patch changelog.

SAVI token — utility token, fees, rewards, staking VOTE token — governance, proportional to SAVI staking 10,000 SAVI staked → 1,000 VOTE (fixed ratio) VOTE is not sellable — it is earned by contributing. Decisions on protocol, treasury, and economic parameters are made by VOTE holders. Nodes with a high PoU score have additional weight on technical proposals.

Savitri Network is the first Layer 1 blockchain where Federated Learning aggregation runs at the consensus layer — not as a smart contract on top, not on an off-chain coordinator. The difference from other "AI + blockchain" projects: · Most projects bolt smart contracts on existing chains → chain only sees a hash of the model → off-chain aggregator is still a single point of trust · Savitri's consensus sees the model weights → aggregation runs IN consensus → no off-chain coordinator to trust The mechanism: the same Proof-of-Unity score that elects block proposers also weights FL gradient updates during the federated round. One mechanism. Two jobs. In practice: testnet live with 45,725 TPS peak / 485ms finality on 25 cloud nodes. 1 enterprise customer in production for 1 year (academic / government deployment). TensorFlow Federated, PyTorch, Flower — all run on top via savitri-fl bridges.

Three problems FL has at production scale that no framework solves: 1. Data Poisoning — bad gradients corrupt the global model; detection requires trust in a reference 2. Free-Rider Problem — without quality-weighted compensation, rational actors free-ride 3. No Provenance — no auditable lineage for compliance (EU AI Act 2026, HIPAA, GDPR) Smart contracts on Ethereum/Solana can't see model weights. Storing a 50MB ResNet update on Ethereum costs ~$10,000 in gas. To do FL aggregation IN consensus, you need new L1 work — not a fork.

TensorFlow Federated, PyTorch, Flower — all via savitri-fl bridges (alpha). Custom FL implementations also supported via direct savitri-fl Rust SDK. You don't rewrite your FL workload. You add 4 API calls: register model, submit gradient, query state, claim quality-weighted rewards. The aggregation runs at consensus layer.

Every node on the network accumulates a "trust score" from 0 to 1000. The score measures: are you online when needed? Are the things you propose valid? Do you participate actively? Nodes with a high score earn more. Nodes with a low score (< 300) receive no rewards — but can get back on track without losing everything. Think of it as an Airbnb rating for your node: the more reliable you are, the more you earn.

No. There is no minimum stake requirement. To run a Lightnode you need: · A laptop or smartphone with 4GB of RAM · A 1 Mbps internet connection · The app or installer (both free) SAVI rewards arrive automatically once your node accumulates a sufficient PoU score (≥ 300/1000).

Yes — it's an architectural choice, not a marketing promise. The Lightnode client is written in Rust with Flutter, compiled natively for ARM64 (iOS and Android). It's not a WebApp in a wrapper — it's a real node. It has three battery consumption modes: Active: full participation (while charging) Balanced: default, reduced heartbeat Power Saving: heartbeat only < 20% battery Runs in the background. Doesn't occupy the screen.

No. Savitri is on testnet. The tokens you earn now are test tokens — they have no economic value and are not transferable. When mainnet goes live (after the audit remediation is complete), that will change. We do not make listing announcements before concrete conditions are in place.

Base: 100 SAVI per epoch. With maximum PoU score (1000/1000): 300 SAVI per epoch (3.0x multiplier). Current epoch: 300 seconds (5 minutes) → ~288 epochs/day Monthly estimate (score 850/1000, uptime 90%): ~250 SAVI/epoch × 0.90 × 288 × 30 ≈ ~1,944,000 SAVI/month Note: approximate calculation on testnet. Use the interactive calculator at /run-node for your specific configuration.

It depends on how long you're offline: Brief interruption (< 1 epoch): Small reduction in availability score. Rewards are reduced proportionally in that epoch. No permanent effect. Extended downtime (days): Score drops by ~10% for each prolonged offline period. You are not excluded from the network — but you earn less until you recover your uptime. The system distinguishes a normal restart from a prolonged absence. Home connection issues are expected and tolerated.

For the Lightnode: no, it's not required. The lightnode connects to the P2P network via Kademlia DHT and maintains connections even with a dynamic IP. For the Masternode: recommended. Masternodes participate in BFT consensus with direct voting — a static IP improves reachability by other masternodes and reduces latency in voting rounds. It's not mandatory, but it improves Availability and Latency scores.

3 options: 1. Installer Wizard — for those who don't want the command line Windows / Linux / macOS — download from /run-node 2. Docker — for those who prefer containers

cd docker && docker-compose up -d
# Starts 2 MN + 2 LN + Grafana in one line

3. Build from source — Rust 1.82+

cargo build --release
# For those who want to audit the code

Full guide: /run-node

Yes, with Docker you can start multiple containers on the same host. The default docker-compose starts 2 masternodes + 2 lightnodes on the same PC — it's the standard testing setup. Each node has: · Separate ed25519 keys (dedicated -keys volume) · Independent PoU score · Separate P2P ports (4001/4002 for LN, 4021/4022 for MN) Note: each node consumes RAM independently. With 4 nodes on a 16GB PC, you have ~4GB per node.

The keys are in the Docker volume (*-keys) or in the installer folder (~/.savitri/keys/ by default). If you back up the keys before losing the PC, you can restore the node on different hardware with the same identity and the same PoU score. If you don't have the backup: you lose the accumulated score and reputation. There is no recovery mechanism — it's by design (security). Seed phrase backup at installation — mandatory.

Not directly EVM-compatible in the sense of executing EVM bytecode. But our smart contract standards are designed to feel EVM-familiar to an Ethereum developer: Savitri-20 → same API as ERC-20 (mint, burn, transfer, approve...) Savitri-721 → same API as ERC-721 Savitri-1155 → same API as ERC-1155 The runtime is Rust-based, not EVM — which means: ✓ Faster ✓ No gas estimation ✗ Solidity bytecode cannot be imported directly ✗ Hardhat/Foundry tools don't connect natively If you use the Savitri SDK, the logic is familiar. The code is not 1:1 portable.

Yes. RPC base URL:

http://testnet.savitrinetwork.com:8546

(or any node running locally) You can use it right now:

curl http://testnet.savitrinetwork.com:8546/savitri/health

Free testnet tokens from the faucet:

POST /savitri/faucet/claim
# 1 claim per address per 24 hours.

docs.savitrinetwork.com — complete documentation: · Consensus spec (PoU, BFT, DAG) · RPC reference (all 13 endpoints) · Rust SDK (savitri-sdk = "0.4") · Flutter SDK (savitri_flutter: ^0.4.0) · Smart contract guide (Savitri-20/721/1155) · Node setup (Docker, installer, source) Quick reference also at /developers — with live playground.

Yes — the repository is public, MIT + Apache 2.0. Contributions welcome via pull request on GitHub. What you need: · CONTRIBUTING.md in the repo for guidelines · Open issue before a large PR · Tests for every new feature Active bug bounty: valid vulnerabilities are rewarded in SAVI. Details at /security.

100,000,000 SAVI. Fixed. Important note: some previous communications indicated "2 billion" — that figure was incorrect. The correct supply hardcoded in the protocol is 100,000,000. Verifiable in the tokens.toml file in the public repository.

No confirmed date. Prerequisites that must be met before listing: 1. Audit remediation complete (0 open critical findings) 2. Post-remediation final audit passed 3. Mainnet live and stable The first planned listing is on native DEX, then at least 1 CEX. Any listing announcement not coming from official @savitrinetwork is unauthorized.

The /team page will be published with verified team data before the mainnet launch. We do not publish partial or unverified information. For institutional investors with serious interest: [email protected] — response within 5 business days.

Yes. Completed in February 2026 by an independent third party. 68 findings (16 critical, 22 high, 18 medium, 12 low). Remediation in progress — ETA Q2 2026. Full details at /security, including the live remediation status and patch changelog.

SAVI token — utility token, fees, rewards, staking VOTE token — governance, proportional to SAVI staking 10,000 SAVI staked → 1,000 VOTE (fixed ratio) VOTE is not sellable — it is earned by contributing. Decisions on protocol, treasury, and economic parameters are made by VOTE holders. Nodes with a high PoU score have additional weight on technical proposals.