Skip to content
Contact us
Developers/Agent Developer Kit

Use private data with AI without exposing it.

Start using customers' private data in agentic workflows safely, with a single prompt.

AICPA SOC 2

The best-in-class AI governance tools, in one prompt.

ISO 27001:2022 Stage 2

Private Data Vault

Allow your agent to work with private data without touching it.

Your AI reasons first, before requesting for private data to be processed in a sealed vault.

  • Encrypt and store your data in a network of hardware sealed TEE enclaves only accessible by you.
  • Results are passed straight to the recognised 3rd party.
  • The AI cannot leak what it does not see or have.

Agent Monitor

Use AI with confidence by knowing what your agent is doing with sensitive data.

Monitor every agent's interaction with private data through tamper-proof logs.

  • Each agent is given an unique DID with hard scoped permissions.
  • Full view of what each agent and their owners have done.
  • The activity log is immutable, essential for checks and audits.
Complying withAICPA SOC 2ISO 27001:2022 Stage 2
How T3 Network works

Separate your AI's thinking from execution.

Your authorised agents ask for sensitive work to be done, work is executed inside a sealed environment, untouchable by the AI.

  1. The agent signs with its own DID did:t3n:agent-7f3a and sends a placeholder like {{user.passport_no}}, never the value.

  2. T3 Network checks the owner's delegation grant: if this agent has the permission to make this request.

  3. The TEE contract reads the encrypted value from the owner's dataset and fills the placeholder in the enclave.

  4. The request goes out from the enclave to the allowlisted API. Any other API gets denied.

  5. The agent gets only the response it needs. The action is written to the tamper-evident log.

Compare

Get cloud scalability with local-level security.

Comparing T3 Network with on-prem local AI and sandboxed AI
T3 NetworkOn-prem local AISandboxed AI
Scales without your own hardwareYes.No. You buy, run, and maintain the GPUsYes.
Works with any modelsYes.You are free to use any best-in-class modelsNo. Only models you can host and compromise withYes.
Private data secure from AI leakYes.AI never touches or sees the dataNot completely. AI still sees raw data locally. The risk is still there if workflows are not isolatedNo. AI still accesses your data in sandbox with a risk of prompt injection
What protects your dataHardware isolation and cryptography. AI cannot leak what it cannot seeYour perimeter and your ops team, prone to errorsContainer or VM isolation from the host, still prone to malicious attacks
Identity and scoped permissions per agentBuilt inNeed resources to build it yourselfNeed resources to build it yourself
Tamper-proof activity logBuilt inNeed resources to build it yourselfSandbox logs only, not tamper-proof.
Cost to set up and maintainFree to start, Bill-as-you-scale (free during launch)Upfront CAPEX with thousands of dollarsPay per sandbox, plus your setup and upkeep
Time to first integrationMinutes, from one promptMonths to years.Days to weeks, per sandbox setup

1 of 3

T3 NetworkSandboxed AIOn-prem local AI
Scales without your own hardwareYes.Yes.No. You buy, run, and maintain the GPUs
Works with any modelsYes.You are free to use any best-in-class modelsYes.No. Only models you can host and compromise with
Private data secure from AI leakYes.AI never touches or sees the dataNo. AI still accesses your data in sandbox with a risk of prompt injectionNot completely. AI still sees raw data locally. The risk is still there if workflows are not isolated
What protects your dataHardware isolation and cryptography. AI cannot leak what it cannot seeContainer or VM isolation from the host, still prone to malicious attacksYour perimeter and your ops team, prone to errors
Identity and scoped permissions per agentBuilt inNeed resources to build it yourselfNeed resources to build it yourself
Tamper-proof activity logBuilt inSandbox logs only, not tamper-proof.Need resources to build it yourself
Cost to set up and maintainFree to start, Bill-as-you-scale (free during launch)Pay per sandbox, plus your setup and upkeepUpfront CAPEX with thousands of dollars
Time to first integrationMinutes, from one promptDays to weeks, per sandbox setupMonths to years.
Pricing

Start free. Pay only for what you store and use.

Every account starts with free credits. In production, TEE execution and storage are metered in T3 Network credits.

Sandbox

Everything you need to build and test.

Freeon testnet
  • 20,000 sandbox credits
  • All features same as production
  • Prompt and skill support for Claude, Codex
  • Developer Telegram support

Pay as you go

First 12 months free

For agents running in production.

Metered usagein T3 Network credits
  • Compute - USD 0.0244 /GB /hour estimated
  • Storage - USD 0.5614 /GB estimated
  • Seamless migration from sandbox to production

Enable high‑stakes autonomous transactions, with low risk, at scalable prices.

Run payroll without exposing employee PII~$1*per payroll cycle for 100 employees
Complete payment without exposing credit card details~$0.45*per transaction
* Costs are estimated and dependent on the complexity of each workflow.

FAQs

Need help? Reach out to us at

devrel@terminal3.io
No. Terminal 3 sits behind your agent's tool calls. Keep Claude, GPT, or any model and framework you already use.
No. The skill file teaches Claude Code or Codex how to call the SDK. Real values are substituted inside the TEE at runtime, so they never enter the assistant's context, chat logs, or your app logs.
It can only reach what was explicitly permissioned to. Any call outside that scope is refused with a denial. Agents will never see the data in plaintext at any point of time nor would they be able to. The agent will not be able to leak what it does not have. All actions will be logged in a tamper-proof activity log, so suspicious actions can be retroactively be observed and flagged.
The data owner. An agent can only authenticate with its own DID, but it cannot make any outbound calls until a user signs a delegation naming its DID.

Fully unlock the value of agentic workflows with a single prompt today.

Sign up for an API key, get free credits on us and paste the setup prompt. It's that simple.

claude
Add Terminal 3 to this project. Install the t3n-adk-quickstart skill from https://docs.terminal3.io/developers/adk/support/ai-coding-assistants, use https://docs.terminal3.io/llms.txt as the docs index, then follow the skill's steps in order and stop once a did:t3n identity prints. Then ask me whether to start with T3 Network Vault (private data out of AI context), T3 Network Monitor (agent identity and activity log), or T3 Network Pay (agent payments within policy).