With the amount of new subnets being added it can be hard to get up to date information across all subnets, so data may be slightly out of date from time to time
DaniLabs is Bittensor Subnet 2. The name stands for Distributed Accelerated Neural Intelligence, and the project’s tagline is “Intelligence, distributed.” DaniLabs is developing an inference subnet: a network focused on running AI models, so that developers and applications can send a request and get a useful response back.
Three design priorities guide how the network and the experience of using it are being built:
DaniLabs took over Subnet 2 on 16 October 2026 under an agreement with Inference Labs, and the subnet’s on-chain name changed to DaniLabs shortly afterwards. The project describes the service as in development. Its first public subnet code, published on 21 October 2026, is validator software: validators are being built to score miners on latency, token throughput, completed requests and availability, so that generation speed alone does not decide a miner’s reward. Since 27 October 2026 the DaniLabs validator has been running in production on the subnet, with miner rewards burned for now while its scoring is verified.
DaniLabs is Bittensor Subnet 2. The name stands for Distributed Accelerated Neural Intelligence, and the project’s tagline is “Intelligence, distributed.” DaniLabs is developing an inference subnet: a network focused on running AI models, so that developers and applications can send a request and get a useful response back.
Three design priorities guide how the network and the experience of using it are being built:
DaniLabs took over Subnet 2 on 16 October 2026 under an agreement with Inference Labs, and the subnet’s on-chain name changed to DaniLabs shortly afterwards. The project describes the service as in development. Its first public subnet code, published on 21 October 2026, is validator software: validators are being built to score miners on latency, token throughput, completed requests and availability, so that generation speed alone does not decide a miner’s reward. Since 27 October 2026 the DaniLabs validator has been running in production on the subnet, with miner rewards burned for now while its scoring is verified.
What exists today
DaniLabs is at an early stage. Its website, danilabs.ai, sets out the idea and the three design priorities, labels the service “in development”, and offers an email sign-up for launch and progress updates. Announcements are posted on X at @danilabsai.
Subnet validator code (public since 21 October 2026)
The subnet repository named in the on-chain identity, github.com/DaniLabsAI/subnet, is now public. It holds the DaniLabs validator software: the code that is to evaluate miner performance, calculate scores and submit weights to Bittensor. According to its README, the validator is being developed to:
Validator reports are meant to carry operational measurements only, without customer prompts or generated responses. The first executable release is a read-only chain observer pilot (commands status, watch and shadow) that checks finalized subnet identities and validator state on a CPU-only host; it opens no wallet and does not submit weights. The scoring formula, eligibility rules and measurement definitions are still being developed and are to be documented and tested in the repository before live weight submission is enabled.
Localnet validator pilot (since 24 October 2026)
The repository now also contains a pilot of how the validator is to be operated, run on a private local development chain. It adds an isolated signer, the only part allowed to set weights and restricted to that development chain and a development key; a supervised worker process that holds no private key; and an automatic updater that installs new worker releases in stages, runs tests and health checks, and rolls a failing update back. In this pilot the validator directs its weight to the subnet owner’s UID and checks that the result settles on chain as expected. Changes are tested on this local chain before they are deployed to Finney, Bittensor’s main network.
Signer key custody and recovery (added 26 October 2026)
The signer now has one fixed transaction policy: it can only set the full weight to the verified subnet owner’s UID, and the worker and updater cannot change the destination, weight, subnet or payload. The validator hotkey is kept as a standard encrypted Bittensor keyfile, and no coldkey private key is needed. Every signer start is locked: the key is held in memory only after a custodian unlocks it on the host. An optional recovery service can unlock the signer automatically after a reboot or crash, using a password stored in Oracle Cloud (OCI) Secret Management, and a persistent emergency lock overrides that recovery.
Production validator on Finney (since 27 October 2026)
On 27 October 2026 DaniLabs announced that its validator switchover was live on Subnet 2. The production validator now runs on Finney in a 100% burn mode: it sets the full weight to the subnet owner’s UID, so miner incentive is burned for now, and it checks separately that the weight row and the burn settle on chain.
Candidate scoring (logged, not yet rewarded)
Once per chain epoch the validator fetches signed performance measurements from the DaniLabs API, verifies the signature against a pinned public key, and checks each miner’s UID, hotkey and registration block against finalized chain state, rejecting stale, duplicate or inconsistent reports. Reports carry no customer prompts or responses. The current scoring policy (performance v2) weights decode speed in tokens per second at 50%, time to first token at 35%, completion reliability at 10% and gateway-to-miner round-trip time at 5%, measured against per-model targets (initially 300 tokens per second and 500 ms to first token for openai/gpt-oss-20b). The resulting candidate weights are logged alongside the burn but are not yet used for rewards: token usage is still reported by the miners themselves, so scores stay ineligible until token counts and correctness can be verified independently.
Other components
The DaniLabs GitHub organisation (github.com/DaniLabsAI) also lists:
The service, platform and miner-kit repositories are not yet publicly visible.
What exists today
DaniLabs is at an early stage. Its website, danilabs.ai, sets out the idea and the three design priorities, labels the service “in development”, and offers an email sign-up for launch and progress updates. Announcements are posted on X at @danilabsai.
Subnet validator code (public since 21 October 2026)
The subnet repository named in the on-chain identity, github.com/DaniLabsAI/subnet, is now public. It holds the DaniLabs validator software: the code that is to evaluate miner performance, calculate scores and submit weights to Bittensor. According to its README, the validator is being developed to:
Validator reports are meant to carry operational measurements only, without customer prompts or generated responses. The first executable release is a read-only chain observer pilot (commands status, watch and shadow) that checks finalized subnet identities and validator state on a CPU-only host; it opens no wallet and does not submit weights. The scoring formula, eligibility rules and measurement definitions are still being developed and are to be documented and tested in the repository before live weight submission is enabled.
Localnet validator pilot (since 24 October 2026)
The repository now also contains a pilot of how the validator is to be operated, run on a private local development chain. It adds an isolated signer, the only part allowed to set weights and restricted to that development chain and a development key; a supervised worker process that holds no private key; and an automatic updater that installs new worker releases in stages, runs tests and health checks, and rolls a failing update back. In this pilot the validator directs its weight to the subnet owner’s UID and checks that the result settles on chain as expected. Changes are tested on this local chain before they are deployed to Finney, Bittensor’s main network.
Signer key custody and recovery (added 26 October 2026)
The signer now has one fixed transaction policy: it can only set the full weight to the verified subnet owner’s UID, and the worker and updater cannot change the destination, weight, subnet or payload. The validator hotkey is kept as a standard encrypted Bittensor keyfile, and no coldkey private key is needed. Every signer start is locked: the key is held in memory only after a custodian unlocks it on the host. An optional recovery service can unlock the signer automatically after a reboot or crash, using a password stored in Oracle Cloud (OCI) Secret Management, and a persistent emergency lock overrides that recovery.
Production validator on Finney (since 27 October 2026)
On 27 October 2026 DaniLabs announced that its validator switchover was live on Subnet 2. The production validator now runs on Finney in a 100% burn mode: it sets the full weight to the subnet owner’s UID, so miner incentive is burned for now, and it checks separately that the weight row and the burn settle on chain.
Candidate scoring (logged, not yet rewarded)
Once per chain epoch the validator fetches signed performance measurements from the DaniLabs API, verifies the signature against a pinned public key, and checks each miner’s UID, hotkey and registration block against finalized chain state, rejecting stale, duplicate or inconsistent reports. Reports carry no customer prompts or responses. The current scoring policy (performance v2) weights decode speed in tokens per second at 50%, time to first token at 35%, completion reliability at 10% and gateway-to-miner round-trip time at 5%, measured against per-model targets (initially 300 tokens per second and 500 ms to first token for openai/gpt-oss-20b). The resulting candidate weights are logged alongside the burn but are not yet used for rewards: token usage is still reported by the miners themselves, so scores stay ineligible until token counts and correctness can be verified independently.
Other components
The DaniLabs GitHub organisation (github.com/DaniLabsAI) also lists:
The service, platform and miner-kit repositories are not yet publicly visible.
Team
DaniLabs is led by Dan (@opendansor on X). According to Inference Labs, Dan joined its team from the Opentensor Foundation (OTF) and his applied research helped scale DSperse, the system previously run on Subnet 2. He describes himself as the architect of DSperse and links its research paper, “DSperse: A Framework for Targeted Verification in Zero-Knowledge Machine Learning” (arXiv 2508.06972). Announcing the move, he wrote that his subnet journey started on SN2 and that he is now leading its next chapter. The on-chain contact is [email protected].
On 27 October 2026 DaniLabs announced that engineer Xen (@xen43 on X) had joined the team to build out its inference platform and infrastructure, with a focus on speed, reliability and keeping costs low.
Subnet ownership
On 16 October 2026 Inference Labs announced that it had entered into an agreement for Dan to take over Subnet 2 as DaniLabs and lead its direction and development, focused on fast, affordable AI inference. The subnet’s registration dates from block 2,734,060 (April 2024) and its owner coldkey (5EcYQ3W77ndrmMWdvVQusoFqY8doxfP3U2zrh7xZQiaz7avY) is unchanged; the on-chain identity was updated to DaniLabs, with the new website, GitHub and X links, on 18 October 2026.
Community
Follow the build on X at @danilabsai and sign up for updates at danilabs.ai. DaniLabs runs its own Discord server (linked above, invite shared by @danilabsai on X on 30 October 2026); general subnet discussion also takes place on the main Bittensor Discord.
Team
DaniLabs is led by Dan (@opendansor on X). According to Inference Labs, Dan joined its team from the Opentensor Foundation (OTF) and his applied research helped scale DSperse, the system previously run on Subnet 2. He describes himself as the architect of DSperse and links its research paper, “DSperse: A Framework for Targeted Verification in Zero-Knowledge Machine Learning” (arXiv 2508.06972). Announcing the move, he wrote that his subnet journey started on SN2 and that he is now leading its next chapter. The on-chain contact is [email protected].
On 27 October 2026 DaniLabs announced that engineer Xen (@xen43 on X) had joined the team to build out its inference platform and infrastructure, with a focus on speed, reliability and keeping costs low.
Subnet ownership
On 16 October 2026 Inference Labs announced that it had entered into an agreement for Dan to take over Subnet 2 as DaniLabs and lead its direction and development, focused on fast, affordable AI inference. The subnet’s registration dates from block 2,734,060 (April 2024) and its owner coldkey (5EcYQ3W77ndrmMWdvVQusoFqY8doxfP3U2zrh7xZQiaz7avY) is unchanged; the on-chain identity was updated to DaniLabs, with the new website, GitHub and X links, on 18 October 2026.
Community
Follow the build on X at @danilabsai and sign up for updates at danilabs.ai. DaniLabs runs its own Discord server (linked above, invite shared by @danilabsai on X on 30 October 2026); general subnet discussion also takes place on the main Bittensor Discord.
DaniLabs has not published a dated roadmap. Its website describes the work in three stages:
Progress so far
DaniLabs has not published a dated roadmap. Its website describes the work in three stages:
Progress so far