Skip to content
TaoLoop
Menu

Private beta · Apex (subnet 1)

Build and evaluate subnet candidates on your VPS.

TaoLoop runs managed coding agents, Codex by default or Claude Code, on a Linux server you control. They write and evaluate Bittensor subnet candidates while every model call is metered against limits you set, and nothing is submitted or paid in TAO without your approval.

Request beta access See the workspace
Apex · SN 1
Illustrative workspace · sample data

Subnet overview

Updated 12 s ago
Next action
Review the evaluated candidate
State
Running
Current work
Coding iteration 2 of 3
AI credits used
0.8 of 2.0 limit
Compute
Your VPS · connected

Agent activity

Codex · live
  1. 14:02Read src/solver.py, tests/test_solver.py
  2. 14:03Ran pytest -q · 14 passed
  3. 14:05Edited src/solver.py (+7 −1)
  4. 14:06Ran pytest -q · 15 passed
  5. 14:06Committed a1c9e2f “Cache candidate scores”
  6. 14:07Queued Local evaluation on this VPS

Code change

src/solver.py
def score(cands, budget):
− out = map(evaluate, cands)
+ cache, out = {}, []
+ for c in cands:
+ k = c.fingerprint()
+ if k not in cache:
+ cache[k] = evaluate(c)
+ out.append(cache[k])
return rank(out, budget)

Committed to your private repository

Evaluation

Local · on your VPS
  • Baseline0.612
  • Candidate a1c9e2f0.631
  • Scenarios8 of 8 completed

Local score only, not a network result. Nothing is submitted until you approve it.

Illustrative workspace with sample data. Panels and labels follow the TaoLoop dashboard; the numbers are not results, rewards or earnings.

PRODUCT

One workspace from first commit to reviewed result.

Each subnet you work on gets a workspace. It shows what is running, what it changed, how the change scored and what it has cost so far.

  • Subnet overview

    The current state of your work on a subnet, what it has used against its limits, and the one thing that needs you next.

  • Live agent activity

    Follow the coding agent as it reads, edits, runs tests and commits on your VPS. Send guidance or stop it at any point.

  • Code changes

    Every change lands as a commit in a private repository you can open in the editor or on GitHub.

  • Evaluation

    Candidates are scored against a pinned benchmark on your VPS. Local scores are labeled as local, never as network results.

HOW IT WORKS

Three steps, with you at the last one.

TaoLoop does the repetitive work. Decisions that cost TAO or change what the network sees stay with you.

  1. 01

    Connect your VPS and repository

    Sign in with GitHub, add a Linux server you control over SSH and pin its host key. TaoLoop installs the runner and creates a private repository for your subnet work.

  2. 02

    Run and evaluate

    Start a run with an objective, an AI credit limit and a time limit. The agent codes on your VPS; each candidate is evaluated there before anything leaves it.

  3. 03

    Review results and the next action

    See the change, its evaluation and what it cost, then decide: iterate, stop, or submit.

When a wallet is needed. Research, coding and local evaluation need only AI credits. Submitting a candidate to a subnet such as Apex needs a registered hotkey and a TAO submission fee, which you approve before anything is signed.

TRUST AND ECONOMICS

Your server, your keys, your limits.

TaoLoop manages the loop; it does not take custody of your compute, your wallet or your spending decisions.

  • Illustration of a private compute node connected to a wider network

    Code runs on your VPS

    Agents, evaluations and miner workloads run in sandboxed containers on the server you connected, not on TaoLoop machines.

  • Illustration of an automated loop passing through control gates

    Keys stay out of the browser

    Model provider keys stay on TaoLoop’s servers and wallet keys stay with an isolated signer. The dashboard never holds either.

  • Illustration balancing compute inputs against a reward signal

    Spending is bounded

    Each run has an AI credit limit it cannot exceed. Every TAO spend is tied to an approval you gave for that exact action.

AI credits

Pay for the agents’ model usage, metered per request through TaoLoop. Each run reserves against its own limit and settles what was actually used.

TAO

Pays network costs such as submission fees, from your wallet, only after you approve the exact spend. Credits and TAO are never mixed.

AVAILABILITY

What is available today

As of October 2026

Access
Private beta, by invitation. Existing beta users sign in with GitHub.
Subnets
Apex (subnet 1) is the supported subnet. Others are researched, not yet run end to end.
Compute
Your own Linux VPS, connected over SSH. Runner installation and workloads run there today.
Coding agents
Codex by default; Claude Code is available as an alternative.
Funding
AI credits are arranged with the TaoLoop team during the beta. Self-serve deposits are not available yet.

Questions

What do I need to get started?

A GitHub account and a Linux x86_64 VPS you control, reachable over SSH with root or passwordless sudo, with systemd and Docker installed. We have tested Ubuntu 24.04 and Debian 12. A wallet is only needed if you decide to submit.

Which coding agents does TaoLoop run?

Codex is the default. Claude Code is available when you start a run or a coding session. Both run natively on your VPS, and their model requests go through TaoLoop’s metered API, so usage is counted against your AI credit limit.

Where does my code live?

In a private repository TaoLoop creates for you in the TaoLoop GitHub organization, with your GitHub account added as a collaborator. Working copies stay on your VPS.

When do I need a wallet or TAO?

Only to submit. Research, coding and local evaluation use AI credits. Submitting to a subnet such as Apex needs a registered hotkey and a TAO fee; TaoLoop prepares the submission, you approve the exact amount, and an isolated signer signs it.

How do approvals and stopping work?

Runs stop at their AI credit and time limits. Spending TAO always waits for your approval of that specific action. You can stop a run or revoke its authority at any time; work already in progress is wound down and nothing new starts.

What is the difference between AI credits and TAO?

AI credits pay for metered model usage by the agents. TAO pays network costs such as submission fees, from your wallet. They are tracked separately and never converted into each other.

What support is there during the beta?

Beta users work directly with the TaoLoop team during onboarding and when something blocks a run. Expect rough edges: we add users in small groups so we can help each one.

Does requesting access create an account?

No. It adds your email to the beta list. We contact you when a place opens and set up your access then.

REQUEST BETA ACCESS

Join the next beta group.

Leave your email and we will contact you when a place opens. We add users in small groups and set up access with each one.

We use your email only to contact you about TaoLoop access.