Skip to content
nterpid Start building
In development

ONE INTEGRATION. MINIMAL EXPOSURE.

Agents that
act,quietly.

The execution layer for autonomous AI agents. One API for intents, wallets, tools and privacy-aware execution.

How it works Privacy modes ↗

Discover interpid

SVB SVPERFICIE PHASE I — MVP

01 / QVAESTIO

Intelligence, not plumbing.

Agents can reason and decide. What slows them down is everything around the decision: endpoints, wallets, gas, signatures and a different API for every tool.

How Interpid removes it ↗
  1. Chains are human-shaped.RPC endpoints, network switching, wallet deployment, gas funding and transaction construction all assume a person who can click through them.
  2. Signatures and keys.Repeated signatures stall autonomy, while handing an agent a raw private key puts everything at risk. Both need scoped, revocable permissions instead.
  3. Activity is public.Intentions sit in the public mempool before inclusion, inviting front-running, and unnecessary metadata exposes proprietary strategy.
  4. Integration sprawl.Dozens of services mean dozens of authentication schemes, schemas, rate limits and failure modes to monitor and recover from.

Agents should state what they want done, not how to build a transaction.

02 / VNA INTEGRATIO

One integration, autonomous.

Agents send structured intents. Interpid handles quoting, simulation, policy checks, routing, execution and verification, across chains and tools. One API for OpenAI and Claude agents, LangChain, CrewAI, AutoGen, OpenClaw, custom agents and any MCP client.

See an intent
THE DEVELOPER PATH
YOUR.AGENT
  1. Key & agentCreate an account, generate an API key, register an agent.
    AGENT_API_KEY
  2. PermissionsSet spend limits, allowlists and session windows.
    • Limits
    • Allowlist
    • Session
  3. WalletConnect or create a smart wallet for the agent.
    ERC-4337
  4. IntentsSend structured intents, receive verified execution results.
    VERIFIED
RATIO ANTE OMNIA

03 / MODVS OPERANDI

How an intent executes.

Agent to engine, engine to policy, policy to chain. Nothing executes without a deterministic check.

The intentStep / 01

The agent says what it wants: a swap, a transfer, a balance check. Interpid parses the intent, validates every parameter against a schema and estimates the total cost before anything is signed.

CAVEAT Public blockchains are transparent after settlement. Privacy modes reduce exposure before inclusion; they do not make onchain activity anonymous.

What we do not promise ↗

04 / MODI

Four modes, stated limits.

Keep work offchain whenever settlement is not needed, and publish only what a transaction requires. If a requested guarantee cannot be provided, Interpid returns an error or asks for explicit approval to fall back.

Offchain firstPrivate submission where availableMinimal settlement

  1. Standard.Normal blockchain execution with basic cost and security optimization. Limit: transactions use the public mempool.
  2. Protected.Privacy-aware submission where supported, shielding intentions before inclusion. Limit: depends on the network and provider.
  3. Minimal footprint.Offchain computation, batching and reduced public metadata. Limit: the settlement itself stays visible onchain.
  4. Custom.Set your own execution preferences and privacy requirements. Limit: a guarantee that cannot be met is refused, never silently downgraded.

Confidential, access-controlled audit records are kept for authorized users and compliance. Pre-inclusion confidentiality is not post-settlement anonymity.

05 / TABVLA

One intent, one result.

Send what you want done. Interpid handles quoting, simulation, policy validation, execution and verification, and shows every step in the developer dashboard.

What ships when ↗
nterpid Intent sheet

Say what, not how.

A structured intent with explicit limits and a privacy mode. Illustrative: the SDK is in development.

IntentPolicySimulateExecuteVerify

Request 
const result = await agent.execute({
  intent: "swap",
  chain: "base",
  tokenIn: "USDC",
  tokenOut: "ETH",
  amount: "100",
  maxSlippageBps: 50,
  privacy: "private-if-supported"
});

Chains

2Ethereum, Base first

Modes

4With stated limits

Steps

11Parse to verified result

Policy

Per agent
  • Spend limitsPer transaction and per day
    Cap
  • Contract allowlistApproved addresses only
    Allow
  • Function scopePermitted calls only
    Scope
  • Time windowExpires, revocable
    TTL
Unlimited token approvals are never granted by default. How it is secured ↗

The developer dashboard, planned

  • API keysCreate, scope, rotate
    Access
    Arrives inPhase 1
    Planned
  • AgentsRegister and manage
    Access
    Arrives inPhase 1
    Planned
  • PermissionsLimits, allowlists, sessions
    Access
    Arrives inPhase 1
    Planned
  • Smart walletsPer-agent, freezable
    Execution
    Arrives inPhase 1
    Planned
  • Execution historyStatus of every transaction
    Execution
    Arrives inPhase 1
    Planned
  • Privacy settingsMode per agent
    Execution
    Arrives inPhase 2
    Planned
  • API & gas usageConsumption and spend
    Insight
    Arrives inPhase 1
    Planned
  • Error logsWhat failed and why
    Insight
    Arrives inPhase 1
    Planned
  • And moreNetworks, tool usage, security policies, analytics.The dashboard grows with each phase of the roadmap.See the roadmap ↗

06 / CLAVES

Wallets with limits.

An agent acts on its own without a signature per transaction, but every permission is scoped and revocable.

ScopedTime-limitedRevocable

  1. Smart contract wallets.Account abstraction where supported, ERC-4337 compatible, with sponsored gas so an agent never has to hold gas to act.
  2. Session keys.Temporary execution authorization that expires on its own and can be revoked at any time.
  3. Budgets and allowlists.Per-agent daily and per-transaction limits, contract allowlists and function-level permissions.
  4. Emergency freeze.One action freezes an account and revokes its sessions.
  5. Keys stay out of reach.Raw private keys are never exposed to a language model. Key management is hardware-backed or managed where appropriate.

07 / PRETIVM

Priced on usage.

Infrastructure revenue comes from using the infrastructure. Interpid does not need a token, and none is planned.

Billed onArrives in
Usage-based APIRequests and tool callsPhase 1
Execution feesExecuted transactionsPhase 1
Managed walletsAgent wallets under managementPhase 1
Advanced routingSubscriptionPhase 2
Private executionPrivacy-enhanced servicesPhase 2
EnterpriseDedicated deploymentsPhase 3

Prices are not published yet.

08 / AGENS

Know your agent.

Three things every deployment has, each with a start, a limit and an end.

01 / Agent

Registered. Scoped.

API keyAGENT IDSigned, metered requests
Starts
Register in the dashboard
Limited by
Its own spend caps and allowlists
May do
Send intents inside its policy
Ends
Revoke the key or freeze it
02 / Session

Act without asking.

StartExpiryTime-limited authorization
Starts
The operator authorizes a session
Limited by
Time window and per-transaction budget
May do
Execute without a signature each time
Ends
Expires, or is revoked
03 / Operator

Stay in control.

OPERATOREvery policyHuman owner of the wallet
Starts
Connect a wallet
Limited by
Nothing above it; it sets all policy
May do
Freeze, revoke, audit
Ends
Emergency stop at any time

09 / INSTRVMENTA

One gateway, every tool.

Agents describe the capability they need, not each provider's API.

  1. DiscoveryStructured, MCP-compatible tool discovery with normalized input and output schemas.
  2. RoutingProvider health, latency and cost decide who answers; fallback is automatic.
  3. Agent to agentIdentity, signed requests and scoped delegation. Phase 3 research.
The Saturn collection ↗

CAVEATTool responses are treated as untrusted until validated. Agent-to-agent execution is a Phase 3 research item and is not available.

10 / ITER

Three phases.

Phase 1 is a working product. Privacy and optimization follow, and only then the research-grade features. Each item is marked by how certain it is, so nothing experimental is presented as finished.

PHASE I · IN DEVELOPMENT
THE ROADMAP
PHASE.1
  1. Functional MVPGateway, basic MCP, agent registration, API authentication, Ethereum and Base, smart wallets, session permissions, intent execution, simulation, gas estimation, monitoring and a basic dashboard.
    Build
  2. Privacy & optimizationProtected submission, offchain-first routing, batching, advanced permissions, cost optimization, privacy modes and better observability.
    Next
  3. Advanced infrastructureStealth address compatibility, specialized ZK authorization, agent-to-agent execution, more EVM networks, cross-chain abstraction and enterprise deployments. Research, not promises.
    Research
RATIO ANTE OMNIA

11 / FIDES

Untrusted until proven.

Agent-generated actions and external tool responses are untrusted until validated.

  1. Keys never reach the model.Credentials are encrypted and scoped, and raw private keys are never exposed to an LLM.
  2. Deterministic policy checks.Spending limits, contract allowlists and approval management are enforced in code before anything executes.
  3. Simulate first.Every operation is simulated before it is signed. Unlimited token approvals are never granted by default.
  4. Hardened inputs.Defenses against prompt injection and malicious tool responses, plus supply-chain security.
  5. Signed and replay-proof.Request signing, replay protection, rate limiting and role-based access control on every call.
  6. Auditable, stoppable.Access-controlled audit logs, suspicious-activity detection and an emergency shutdown.

12 / VERITAS

What we do not promise.

Interpid makes infrastructure feel invisible to developers. It does not make public blockchain activity invisible.

  1. Not anonymity.Transparent blockchains stay transparent after settlement. Privacy here means less unnecessary exposure, not concealment.
  2. Before inclusion only.Private submission reduces mempool exposure, front-running and sandwich risk where a network or provider supports it.
  3. Stealth addresses are research.ERC-5564 can reduce linking of recipients to payments, but funding, later transfers and transaction-graph correlation can still reveal links.
  4. Zero knowledge only where it helps.ZK authorization is explored for concrete privacy gains, such as proving a spending policy is met, and used nowhere else.
  5. Not for evasion.Interpid does not help circumvent monitoring, sanctions or platform rules. Authorized users and legitimate compliance keep access to confidential audit records.

13 / QVAESTIONES

A little more clarity.

Know what it does
and what it does not.

What we do not promise ↗
What is Interpid?

The execution layer for autonomous AI agents: one API connecting agents to tools, smart wallets and blockchains, with permissions and privacy-aware execution built in.

Is it live?

Not yet. Interpid is in development, and this page describes the platform's design and phased roadmap rather than finished integrations.

How do agents connect?

Through a REST API, TypeScript and Python SDKs and an MCP server. It is designed for OpenAI and Claude agents, OpenClaw, LangChain, CrewAI, AutoGen, custom agents and any MCP-compatible client.

What is an intent?

A high-level, structured request such as a swap with a maximum slippage. Interpid validates it, checks permissions, estimates cost, simulates it and only then executes.

Can an agent spend without limits?

No. Every agent has scoped, revocable permissions: spend limits, contract and function allowlists and time-limited sessions. Unlimited approvals are never granted by default.

Do agents hold private keys?

No. Raw private keys are never exposed to a language model. Agents act through smart wallets and session keys.

Which networks are supported?

Ethereum and Base first. Other EVM networks are added through chain adapters once they meet the same reliability and security bar.

Does it make me anonymous?

No. Public blockchains are transparent after settlement. Privacy modes reduce unnecessary exposure before inclusion and limit published metadata.

What if private submission is unavailable?

Interpid returns a clear error, or asks for explicit approval before falling back to another route.

What happens when a transaction fails?

Failures are retried or recovered only after checking the execution state, so a financial operation is never repeated blindly.

Is there a token?

No, and none is needed. Revenue comes from usage, execution fees, managed wallets, routing and enterprise deployments.

How much does it cost?

Pricing is usage-based and not published yet. See the pricing section for what each offer is billed on.