> For the complete documentation index, see [llms.txt](https://docs.thetpa.ai/deai-tokenized-processing-architecture/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.thetpa.ai/deai-tokenized-processing-architecture/tpai-chain-full-architecture/the-tpa-the-gateway-to-ai-power.md).

# The TPA: the gateway to AI power

A TPA is an AI endpoint on TPAI Chain, and it is the only way to reach the GPU pool.

<figure><img src="https://2937558112-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F7bJZTBY9aHXlYfgnTW9F%2Fuploads%2FmhsDUgzGQpEVdPfnIVyd%2Ftpa-flow.png?alt=media" alt="TPA architecture flow · earlier TPCOIN.AI diagram, updated for TPAI Chain"><figcaption><p>TPA architecture flow · earlier TPCOIN.AI diagram, updated for TPAI Chain</p></figcaption></figure>

The user picks a model or agent by its AI address. The TPA's AI contract converts TPC into AI usage at a US dollar price, routes the job, and settles the split on chain.

|                      | GPU pool TPA                                  | External API TPA                                                |
| -------------------- | --------------------------------------------- | --------------------------------------------------------------- |
| Where the model runs | Our GPU pool                                  | The TPA owner's own API (for example a frontier model provider) |
| Who sets the price   | The price formula: GPU cost ÷ 20%             | The TPA owner                                                   |
| Split                | 60% TPA owner · 20% GPU cost · 20% validators | 80% TPA owner · 20% validators                                  |
| Proof                | Signed receipt plus proof of compute          | Receipt signed by the owner, labelled External                  |
| Privacy levels       | Standard or Confidential                      | External only, never Confidential                               |

**AI contracts live inside a TPA.** An AI contract holds the rules for a model or agent: who can call it, its price, its permissions and where its income goes. The model itself runs on the GPU pool or the outside API and sends back a receipt.

**Agents can work together.** An AI contract can call and pay AI contracts in other TPAs: a legal agent can hire a translation agent. Agents can hold and spend TPC within limits the TPA owner sets: a daily spending cap, approved contracts only, and approval keys for anything above the cap.

**Trained models stay home.** A model fine-tuned on the GPU pool can only run on TPAI infrastructure. Its weights are encrypted, bound to one TPA, and move only when that TPA is transferred. Export is by enterprise contract only at launch. Enterprises that fine-tune on their own data can export their model under an enterprise contract from launch; everyone else's models stay on TPAI Chain. Regulated customers can also keep data and compute on Canadian GPUs.

**External API owners are accountable.** Anyone can connect an outside API to a TPA, with KYC, a declared model and a slashable bond. External API TPAs are trusted reporting backed by accountability: they are labelled External, validators run quality checks and can remove the TPA and slash the bond on evidence of misreporting, and they are never used for Confidential work. Official model makers get a Verified provider badge.

## TPA addresses and the TPA network

**A TPA address is three letters and three digits**, like ABC-042. It is the address of an AI endpoint, not a wallet: wallets use TPAI21- addresses controlled by two or more private keys.

**Why this format.**

* **Readable:** seven characters, easy to say, type and remember, like a licence plate.
* **Compact:** 26 × 26 × 26 letters × 10 × 10 × 10 digits = 17,576,000 addresses, a number that fits in 25 bits.
* **No DNS:** a fixed formula turns a TPA address into its network IP, so finding a TPA needs no name server. The private network routes traffic to wherever that TPA runs, and every connection is checked against the TPA's verified identity, not just its IP.
* **A fixed pool:** 17,576,000 is the total. Addresses of deactivated TPAs are freed and reused by new owners after a cooling-off period, so nobody can pose as the old TPA.

<figure><img src="https://2937558112-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F7bJZTBY9aHXlYfgnTW9F%2Fuploads%2FXiB5C27Cd1EUbUKZMAkw%2Ftpa-network.png?alt=media" alt="TPA addresses and the TPA network · address to IP, reserved range, safety controls"><figcaption><p>TPA addresses and the TPA network · address to IP, reserved range, safety controls</p></figcaption></figure>

**Each TPA works like a Kubernetes pod \[Preferred].** It is isolated, scalable and has its own IP in a private /8 overlay network (10.0.0.0/8, RFC 1918), the range Kubernetes commonly uses for pods. A /8 network holds 2²⁴ = 16,777,216 addresses.

**798,784 addresses are reserved.** 17,576,000 − 16,777,216 = 798,784. Addresses AAA-000 to YVH-215 map one-to-one to the /8 network; the range YVH-216 to ZZZ-999 is reserved for base models and network needs.

**The chain decides, the network enforces.** VXLAN gives up to 16,777,216 isolated network segments (classic VLANs allow only 4,094), so every TPA can have its own segment \[Preferred]. By default every TPA runs in its own isolated network segment; an owner can place several of their TPAs in a shared segment when they need to work together, and validators can quarantine any TPA into isolation at any time. The fit is exact: of the 17,576,000 addresses, the first 16,777,216 map one-to-one to the 16,777,216 IPs of the /8 network and to the 16,777,216 VXLAN segments (24 bits), and 798,784 are reserved. With network policies, validators can quarantine a TPA, block traffic between TPAs, or isolate a group, for example every TPA running a frontier Stage model.

**Clarifications.**

* The limit is 16,777,216 TPA endpoints, not agents: many agents and dApps can live inside one TPA.
* Validators decide; datacenters enforce. Bonded datacenters are contractually required to apply the chain's network decisions.
* For an external API TPA, quarantine cuts TPAI's access to that model, but it cannot stop the provider's model itself.

The overlay network is private, inside TPAI infrastructure, not the public internet.

## TPA subscription

Every TPA pays a flat subscription of $10 a month, priced in US dollars and paid in TPC, whatever the TPA cost. The subscription starts on TPAI Chain at mainnet (11 October 2027); none is charged on Ethereum. TPAs bought before 29 March 2027 get 12 months free after mainnet, so their first payment is due in October 2028. An owner who misses a payment first gets a grace period. Then the TPA is put on hold and can be recovered by paying what is owed plus a recovery fee. If it is never recovered, the owner loses the TPA and everything bound to it (fine-tuned models, agents and data), as with a domain name, and its address returns to the pool after the cooling-off period. The owner's wallet and TPC are not affected, since a TPA is not a wallet. The exact periods and the recovery fee are an open item.

The fee splits 50/50: half to validators, which gives nodes a steady income from day one, and half to Community & Ecosystem, which refills the pool that pays staking and PING rewards. Stage takes no protocol fee.

## Meta Energy: PING

When someone mints a new TPA, they can **ping** an existing TPA. PING is a direct, single-level referral: only the pinged TPA is rewarded, with nothing from referrals of referrals. Self-ping is allowed: an owner can buy more TPAs and ping their own first TPA. Self-ping cannot drain the pool: because the Meta Energy Level starts at 17.576%, an owner who self-pings from the start gets back at most about 55% of what they spend. Each TPA's ping count is stored in its metadata.

**The base unit.** α (ALPHA) = Meta Energy Level × TPA price. The Meta Energy Level starts at 17.576% and only falls, toward zero, as more TPAs are sold: once TPA is a known brand, large referral rewards are no longer needed. The level is set to keep the Meta Energy budget balanced: until mainnet Stage sets it (it can only go down, never up, and every change is published 30 days in advance); from mainnet Stage proposes changes and the 1,000 validators approve them by a 2/3 vote.

| Energy      | Amount            | When                                                                                     | Paid to                 |
| ----------- | ----------------- | ---------------------------------------------------------------------------------------- | ----------------------- |
| δ DELTA     | 0.1 α             | The pinging owner's first purchase; ends after the first ALPHA                           | The pinging TPA's owner |
| α ALPHA     | 1 α               | Every ping; never ends                                                                   | The pinged TPA          |
| β BETA      | 2 α               | The first ping received                                                                  | The pinged TPA's owner  |
| γ GAMMA     | 3 α               | The 5th ping                                                                             | The pinged TPA's owner  |
| λ LAMBDA    | 10 α              | Each 10th ping, up to the 90th (λ1 to λ9, 90 α in total); ends at the first OMEGA        | The pinged TPA's owner  |
| ω OMEGA     | 100 α             | Each 100th ping, up to the 900th (ω1 to ω9, 900 α in total); ends at the first OMEGA MAX | The pinged TPA's owner  |
| ω MAX       | 1,000 α           | Each 1,000th ping                                                                        | The pinged TPA's owner  |
| After ω MAX | 2.9741 α per ping | Every ping after the 1,000th, with no end date                                           | The pinged TPA's owner  |

The order is ω > λ > γ > β > α > δ > 0. The bonus energies (δ, β, γ, λ, ω) stop at ω MAX. After that, each ping after the 1,000th earns 2.9741α, with no end date, at the current Meta Energy Level and within the reward cap, and every reward shrinks as the Meta Energy Level falls toward zero. 2.9741α is a fixed rate set by the Meta Energy formula; the full calculation will be published later.

**Trusted minter status.** There are no paid levels after ω MAX. Status levels recognise the biggest sellers at no extra cost to Community & Ecosystem, and all three are trusted minters.

| Pings   | Status         | What it unlocks                                                         |
| ------- | -------------- | ----------------------------------------------------------------------- |
| 1,000   | ω MAX          | 2.9741α per ping, with no end date; trusted minter badge on the TPA     |
| 10,000  | Gold Minter    | Trusted minter badge; featured in the TPA directory                     |
| 100,000 | Diamond Minter | Verified minter-branch status for brokers; first look at new TPA ranges |

**What levels unlock.** Each level unlocks higher TPC rewards and stronger compute priority: compute priority follows the same Meta Energy formula, so a TPA with more PING gets better access to the GPU pool when it is busy. GPU power goes to people who pay for it: the formula sets the order only among paying jobs, and the fair-share cap still applies, so no TPA can push others out.

**Who pays.** PING rewards come from Community & Ecosystem, never from the mint payment. Before vesting starts on 11 October 2028, staking rewards and PING rewards together can never exceed 49% of Community & Ecosystem (1,722,448,000 TPC); when the cap is reached, rewards stop until the next budget period.

**Minter branches.** Brokers and companies use Meta Energy at scale: they unlock LAMBDA and OMEGA rewards, buy TPAs at a lower effective cost and resell them on their own websites, through Stage's ready API, like domain registrars. Every resold TPA still pays the monthly subscription.
