Subnet Directory

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of new subnets being added it can be hard to get up to date information across all subnets, so data may be slightly out of date from time to time

Subnet 79

τaos

Emissions
Value
Recycled
Value
Recycled (24h)
Value
Registration Cost
Value
Active Validators
Value
Active Miners
Value
Active Dual Miners/Validators
Value

ABOUT

What exactly does it do?

τaos is a specialized Bittensor network for simulating automated trading and financial markets. Its mission is to incentivize risk‑managed AI trading algorithms by running large-scale market simulations. Miners act as trading agents in parallel simulated orderbooks, producing realistic limit-order-book (LOB) data and adaptive strategies. The subnet’s goal is to generate high-quality datasets and algorithms that can outperform others in risk‑adjusted returns. This is accomplished by allowing agents full access to deep limit-order-book (dLOB) data, enabling the development of “deeply intelligent” high-frequency trading strategies. τaos positions itself as a decentralized trading simulator within Bittensor, serving use cases such as:

  • Real-world trading candidates: The best-performing subnet algorithms could be deployed on real exchanges (e.g. market-making) once higher-frequency actions are supported.
  • Rich data for AI/finance: It produces realistic deep-LOB data useful for AI development and quantitative finance (e.g. option pricing).
  • Exchange and regulatory testing: Crypto exchanges or regulators could use τaos to experiment with market designs (tick sizes, fee structures) and measure market-quality metrics (volatility, spreads, crash risk) under various rules.
  • Risk modeling (“Matrix” concept): Running thousands of independent simulations (a “Monte Carlo of markets”) improves statistical accuracy and exposes tail risks in ways standard finance often misses.

 

Overall, τaos provides a Bittensor subnet devoted to financial risk management and trading AI, expanding the ecosystem into decentralized finance simulations.

 

τaos is a specialized Bittensor network for simulating automated trading and financial markets. Its mission is to incentivize risk‑managed AI trading algorithms by running large-scale market simulations. Miners act as trading agents in parallel simulated orderbooks, producing realistic limit-order-book (LOB) data and adaptive strategies. The subnet’s goal is to generate high-quality datasets and algorithms that can outperform others in risk‑adjusted returns. This is accomplished by allowing agents full access to deep limit-order-book (dLOB) data, enabling the development of “deeply intelligent” high-frequency trading strategies. τaos positions itself as a decentralized trading simulator within Bittensor, serving use cases such as:

  • Real-world trading candidates: The best-performing subnet algorithms could be deployed on real exchanges (e.g. market-making) once higher-frequency actions are supported.
  • Rich data for AI/finance: It produces realistic deep-LOB data useful for AI development and quantitative finance (e.g. option pricing).
  • Exchange and regulatory testing: Crypto exchanges or regulators could use τaos to experiment with market designs (tick sizes, fee structures) and measure market-quality metrics (volatility, spreads, crash risk) under various rules.
  • Risk modeling (“Matrix” concept): Running thousands of independent simulations (a “Monte Carlo of markets”) improves statistical accuracy and exposes tail risks in ways standard finance often misses.

 

Overall, τaos provides a Bittensor subnet devoted to financial risk management and trading AI, expanding the ecosystem into decentralized finance simulations.

 

PURPOSE

What exactly is the 'product/build'?

The τaos subnet uses Bittensor’s incentive framework with specialized roles and metrics:

  • Subnet Owner: Designs and tunes the simulation parameters, logic, and reward metrics. The owner ensures fair operation and selects performance metrics (like risk-adjusted returns) so that agents are motivated to behave honestly and produce valuable, research-grade outputs.
  • Validators: Run two linked components – a C++ simulator and a Python validator script – to maintain the simulation state and reward agents. The C++ engine (based on the MAXE framework) handles market microstructure and orderbook matching. The Python validator connects this engine to the Bittensor network: it publishes state updates to miners, collects their responses, computes scores, sets neuron weights, and issues rewards.
  • Miners (Agents): Act as trading bots. Each miner hosts a strategy that observes the current market state and submits instructions (e.g. place/cancel orders) back to the simulator. The miner’s objective is to maximize its risk-adjusted performance (currently the intraday Sharpe ratio of inventory value) across all simulations, while meeting minimum trading volume requirements. This encourages active, algorithmic trading instead of passive strategies. (Simple example agents are provided, but miners are expected to develop custom logic to compete.)

 

Rewards are allocated based on performance. Validators score miners using a weighted sum of risk metrics; initially only intraday Sharpe is used, plus a trading-volume cutoff. Miners below the volume threshold only earn partial reward, ensuring that bots trade frequently. Agents are also incentivized to respond quickly: when the simulator pauses to collect actions, a miner’s latency (response time) is recorded, and slower responses face more price slippage in the resumed simulation. This rewards both intelligent strategies and computational speed. In effect, every agent’s orders are fed into a central matching engine alongside “background” agents, so market impact and liquidity effects are fully accounted for.

 

Simulation Engine and Architecture

Technically, τaos runs a distributed, agent-based market simulator:

  • C++ Simulator: The core is a high-performance C++ engine (with Python bindings) that builds full L3 order books for multiple assets. It faithfully reproduces orderbook microstructure (matching, book state) at high frequency. Background (noise) agents create baseline market conditions (based on established models), while miners’ orders are injected as if on a real exchange. The simulator can run many parallel orderbook realizations to ensure statistical significance.
  • Python Validator: Acts as a bridge to Bittensor. It initializes and coordinates the C++ simulator, communicates state to miners via subtensor requests, receives their order instructions, applies them to the simulator, then computes scores and submits results to the chain. The validator handles all chain API interactions (authentication, weight-setting).
  • Miner Software: Each miner runs the τaos miner client (Python) which receives periodic state updates and returns trade actions. The miner logic (users’ custom AI/trading code) is pluggable: strategies are defined in separate “agent” classes. Miners connect via the usual Bittensor axon interface to participate in this subnet. The miner software supports multiple environment setup (prometheus-node-exporter, nvm/pm2, tmux, etc.) as detailed in the repo.

 

The infrastructure stack requires modern hardware. Validators currently need ~16 GB RAM, 8-core CPU (Ubuntu 22.04+), C++14 compiler, CMake, etc. (Future plans aim to scale from the initial 10 orderbooks up to hundreds or thousands.) Miners have no strict requirements beyond a typical Bittensor node; they chiefly need enough CPU/memory to run their strategy code efficiently. Interaction with Bittensor’s chain: The subnet runs on Bittensor MainNet (tested via entry points). The validator sets neuron weights corresponding to miner performance, so miners earn TAO rewards according to Bittensor’s protocol. (Future upgrades to the Bittensor economy – e.g. dynamic TAO “dTAO” and subnet tokenization – will apply to τaos like other subnets.)

 

Ecosystem and Tools

τaos provides several tangible resources and interfaces:

  • Official Website and Documentation: The project website taos.im hosts an overview, whitepaper links, goal/idea pages, and team info. The whitepaper (linked from the site) details the theoretical design and potential applications. The site also links to social channels and a live monitoring dashboard.
  • Code Repositories: All source code is open on GitHub under taos-im/sn-79. This includes the C++ simulator (simulate/trading/), the Python validator and miner (taos/im/neurons/), example agent strategies, installation scripts, etc. The README contains detailed instructions for installation and running miners/validators.
  • Dashboard/Monitor: A live web dashboard (at taos.simulate.trading) visualizes subnet activity and metrics in real time. It tracks ongoing simulations, agent scores, and other statistics.
  • Community Channels: The team maintains a Discord server (invite via the website) and social media. The Twitter/X account @taos_im (tauos.im) and a Bluesky handle (taos-im.bsky.social) post updates. A Medium blog (e.g. articles by “7/.taos.im”) covers technical topics. These channels share progress, such as the subnet launch and research insights.
  • Backprop Terminal: (Third-party) Tools like Backprop’s dTAO Terminal support buying/selling and staking the subnet’s dynamic TAO (“alpha”) tokens once dTAO features are enabled. For example, SN-79’s alpha token is listed as “SN-79 TAOS” on Backprop, allowing users to trade or stake into the subnet’s pools. (This aligns τaos with Bittensor’s broader subnet tokenomics.)

 

The τaos subnet uses Bittensor’s incentive framework with specialized roles and metrics:

  • Subnet Owner: Designs and tunes the simulation parameters, logic, and reward metrics. The owner ensures fair operation and selects performance metrics (like risk-adjusted returns) so that agents are motivated to behave honestly and produce valuable, research-grade outputs.
  • Validators: Run two linked components – a C++ simulator and a Python validator script – to maintain the simulation state and reward agents. The C++ engine (based on the MAXE framework) handles market microstructure and orderbook matching. The Python validator connects this engine to the Bittensor network: it publishes state updates to miners, collects their responses, computes scores, sets neuron weights, and issues rewards.
  • Miners (Agents): Act as trading bots. Each miner hosts a strategy that observes the current market state and submits instructions (e.g. place/cancel orders) back to the simulator. The miner’s objective is to maximize its risk-adjusted performance (currently the intraday Sharpe ratio of inventory value) across all simulations, while meeting minimum trading volume requirements. This encourages active, algorithmic trading instead of passive strategies. (Simple example agents are provided, but miners are expected to develop custom logic to compete.)

 

Rewards are allocated based on performance. Validators score miners using a weighted sum of risk metrics; initially only intraday Sharpe is used, plus a trading-volume cutoff. Miners below the volume threshold only earn partial reward, ensuring that bots trade frequently. Agents are also incentivized to respond quickly: when the simulator pauses to collect actions, a miner’s latency (response time) is recorded, and slower responses face more price slippage in the resumed simulation. This rewards both intelligent strategies and computational speed. In effect, every agent’s orders are fed into a central matching engine alongside “background” agents, so market impact and liquidity effects are fully accounted for.

 

Simulation Engine and Architecture

Technically, τaos runs a distributed, agent-based market simulator:

  • C++ Simulator: The core is a high-performance C++ engine (with Python bindings) that builds full L3 order books for multiple assets. It faithfully reproduces orderbook microstructure (matching, book state) at high frequency. Background (noise) agents create baseline market conditions (based on established models), while miners’ orders are injected as if on a real exchange. The simulator can run many parallel orderbook realizations to ensure statistical significance.
  • Python Validator: Acts as a bridge to Bittensor. It initializes and coordinates the C++ simulator, communicates state to miners via subtensor requests, receives their order instructions, applies them to the simulator, then computes scores and submits results to the chain. The validator handles all chain API interactions (authentication, weight-setting).
  • Miner Software: Each miner runs the τaos miner client (Python) which receives periodic state updates and returns trade actions. The miner logic (users’ custom AI/trading code) is pluggable: strategies are defined in separate “agent” classes. Miners connect via the usual Bittensor axon interface to participate in this subnet. The miner software supports multiple environment setup (prometheus-node-exporter, nvm/pm2, tmux, etc.) as detailed in the repo.

 

The infrastructure stack requires modern hardware. Validators currently need ~16 GB RAM, 8-core CPU (Ubuntu 22.04+), C++14 compiler, CMake, etc. (Future plans aim to scale from the initial 10 orderbooks up to hundreds or thousands.) Miners have no strict requirements beyond a typical Bittensor node; they chiefly need enough CPU/memory to run their strategy code efficiently. Interaction with Bittensor’s chain: The subnet runs on Bittensor MainNet (tested via entry points). The validator sets neuron weights corresponding to miner performance, so miners earn TAO rewards according to Bittensor’s protocol. (Future upgrades to the Bittensor economy – e.g. dynamic TAO “dTAO” and subnet tokenization – will apply to τaos like other subnets.)

 

Ecosystem and Tools

τaos provides several tangible resources and interfaces:

  • Official Website and Documentation: The project website taos.im hosts an overview, whitepaper links, goal/idea pages, and team info. The whitepaper (linked from the site) details the theoretical design and potential applications. The site also links to social channels and a live monitoring dashboard.
  • Code Repositories: All source code is open on GitHub under taos-im/sn-79. This includes the C++ simulator (simulate/trading/), the Python validator and miner (taos/im/neurons/), example agent strategies, installation scripts, etc. The README contains detailed instructions for installation and running miners/validators.
  • Dashboard/Monitor: A live web dashboard (at taos.simulate.trading) visualizes subnet activity and metrics in real time. It tracks ongoing simulations, agent scores, and other statistics.
  • Community Channels: The team maintains a Discord server (invite via the website) and social media. The Twitter/X account @taos_im (tauos.im) and a Bluesky handle (taos-im.bsky.social) post updates. A Medium blog (e.g. articles by “7/.taos.im”) covers technical topics. These channels share progress, such as the subnet launch and research insights.
  • Backprop Terminal: (Third-party) Tools like Backprop’s dTAO Terminal support buying/selling and staking the subnet’s dynamic TAO (“alpha”) tokens once dTAO features are enabled. For example, SN-79’s alpha token is listed as “SN-79 TAOS” on Backprop, allowing users to trade or stake into the subnet’s pools. (This aligns τaos with Bittensor’s broader subnet tokenomics.)

 

WHO

Team Info

τaos is developed by a core team with deep expertise in high-frequency trading data, market simulations, and AI. The team members remain pseudonymous, but public descriptions highlight that they include:

  • C++ systems engineers focused on ultra-low-latency simulation and validator efficiency.
  • Researchers and quant traders with decades of experience in agent-based market modeling and algorithmic trading.
  • AI/ML experts to integrate deep-learning into trading agents.

 

The team emphasizes that they have “20+ years experience in high-frequency data” and multiple members with PhDs, investing heavily in realistic liquidity modeling. They leverage prior industry work on limit-order-book data, applying it here for the first time on-chain. Core contributors maintain an active presence under the project’s brand: the Twitter/X account (@taos_im) and Medium handle (“7/.taos.im”) are operated by the development group. The GitHub organization taos-im contains their code and documentation. In community forums (Discord, Bittensor Reddit/X), team members occasionally post updates, but they largely speak through official channels. (For example, the “WE ARE LIVE!” blog post on May 7, 2025 announced the mainnet launch.)

 

τaos is developed by a core team with deep expertise in high-frequency trading data, market simulations, and AI. The team members remain pseudonymous, but public descriptions highlight that they include:

  • C++ systems engineers focused on ultra-low-latency simulation and validator efficiency.
  • Researchers and quant traders with decades of experience in agent-based market modeling and algorithmic trading.
  • AI/ML experts to integrate deep-learning into trading agents.

 

The team emphasizes that they have “20+ years experience in high-frequency data” and multiple members with PhDs, investing heavily in realistic liquidity modeling. They leverage prior industry work on limit-order-book data, applying it here for the first time on-chain. Core contributors maintain an active presence under the project’s brand: the Twitter/X account (@taos_im) and Medium handle (“7/.taos.im”) are operated by the development group. The GitHub organization taos-im contains their code and documentation. In community forums (Discord, Bittensor Reddit/X), team members occasionally post updates, but they largely speak through official channels. (For example, the “WE ARE LIVE!” blog post on May 7, 2025 announced the mainnet launch.)

 

FUTURE

Roadmap

Key milestones so far:

Whitepaper & Testnet (Early 2025): The team published a detailed whitepaper (April 2025) outlining τaos’s design. Early testing and development occurred through Spring 2025.

Mainnet Launch (May 7, 2025): After extensive preparation, τaos went live on Bittensor MainNet. The inaugural simulations began, and mining was enabled. (Announcement: “On 7th of May, 2025, the subnet known as TAOS (Sn-79) finally went live on MainNet Bittensor!”.)

Initial Operation: Early runs are using a limited setup (e.g. 10 simulated orderbooks with ~1000 background agents). Validator hardware and parameters have been tuned to ensure stable operation and consensus.

 

Ongoing and future plans include:

  • Scaling Simulations: Gradually increasing the number of parallel orderbooks (target ~1,000+) and background agents to improve statistical validity. This may require more powerful validators or distributed validator pools.
  • Docker Deployment: Simplify deployment by providing Docker images for validator/miner. A Dockerized release is “coming soon” to streamline setup.
  • Enhanced Agent Support: Expanding the library of example agents and tools for common strategy classes, to help miners get started. (Currently only basic random-order agents are included.)
  • Continuous Streaming: Moving from the current discrete update cycle to continuous bidirectional state updates, enabling true real-time high-frequency interaction.
  • Advanced Metrics: Incorporating additional performance and risk measures beyond Sharpe (e.g. drawdown, information ratio) as needed to refine competition. The incentive formula remains under review to ensure data utility.
  • Realism and Events: Introducing more realistic exchange features and external events (adaptive tick sizes, news shocks, regulatory rules) over time, further closing the gap with actual markets. This follows the project’s vision of a “market matrix” that can include exogenous shocks.
  • Ecosystem Integration: Collaborating with Bittensor’s ecosystem, including eventual support for dynamic TAO (dTAO) staking, subnet “alpha” tokens, and interoperability with other subnets and applications.

 

In summary, Subnet 79 τaos is a live, evolving simulation platform in the Bittensor network, launched May 2025, and actively under development. Its road map emphasizes rapid feature iteration and scaling: the team regularly publishes updates as new simulation configurations go live each week. The project’s open-source nature and community channels reflect an ongoing commitment to transparency and collaboration.

 

Key milestones so far:

Whitepaper & Testnet (Early 2025): The team published a detailed whitepaper (April 2025) outlining τaos’s design. Early testing and development occurred through Spring 2025.

Mainnet Launch (May 7, 2025): After extensive preparation, τaos went live on Bittensor MainNet. The inaugural simulations began, and mining was enabled. (Announcement: “On 7th of May, 2025, the subnet known as TAOS (Sn-79) finally went live on MainNet Bittensor!”.)

Initial Operation: Early runs are using a limited setup (e.g. 10 simulated orderbooks with ~1000 background agents). Validator hardware and parameters have been tuned to ensure stable operation and consensus.

 

Ongoing and future plans include:

  • Scaling Simulations: Gradually increasing the number of parallel orderbooks (target ~1,000+) and background agents to improve statistical validity. This may require more powerful validators or distributed validator pools.
  • Docker Deployment: Simplify deployment by providing Docker images for validator/miner. A Dockerized release is “coming soon” to streamline setup.
  • Enhanced Agent Support: Expanding the library of example agents and tools for common strategy classes, to help miners get started. (Currently only basic random-order agents are included.)
  • Continuous Streaming: Moving from the current discrete update cycle to continuous bidirectional state updates, enabling true real-time high-frequency interaction.
  • Advanced Metrics: Incorporating additional performance and risk measures beyond Sharpe (e.g. drawdown, information ratio) as needed to refine competition. The incentive formula remains under review to ensure data utility.
  • Realism and Events: Introducing more realistic exchange features and external events (adaptive tick sizes, news shocks, regulatory rules) over time, further closing the gap with actual markets. This follows the project’s vision of a “market matrix” that can include exogenous shocks.
  • Ecosystem Integration: Collaborating with Bittensor’s ecosystem, including eventual support for dynamic TAO (dTAO) staking, subnet “alpha” tokens, and interoperability with other subnets and applications.

 

In summary, Subnet 79 τaos is a live, evolving simulation platform in the Bittensor network, launched May 2025, and actively under development. Its road map emphasizes rapid feature iteration and scaling: the team regularly publishes updates as new simulation configurations go live each week. The project’s open-source nature and community channels reflect an ongoing commitment to transparency and collaboration.

 

NEWS

Announcements

τaos Follow 83 275

Decentralized Simulation of Automated Trading in Intelligent Markets

taos_im
Load More

View other subnets

Back to the Directory