Agents

How PocketAgent agents are configured, funded, and authorized to use blockchain tools.

An agent is an isolated identity with its own multi-chain wallets, capability set, spending caps, and access token. The LLM orchestrator only invokes tools the agent is allowed to use.

Core fields

FieldDescription
nameDisplay name
chainsSubset of 52 supported chain keys the agent may target
capabilitiesTool capability groups: read, compare, transact, analytics
spending_capPer-chain notional spend limit enforced before broadcast
wallet_addressesDerived addresses for EVM, Solana, SUI, NEAR, Cosmos, TRON

Capability groups

  • read — Balance queries, blocks, transactions, token metadata. Always available when enabled.
  • compare — Gas comparison, chain recommendation, cost estimation.
  • transact — Sends, token transfers, contract writes, simulation. Gated by caps and balance.
  • analytics — Relay stats, history, and cost breakdown for the agent.

Multi-chain wallets

On creation, the platform generates protocol-specific key material, encrypts it with ENCRYPTION_KEY, and exposes public addresses per chain family. Write tools sign with the appropriate encrypted key for the target protocol.

Soft delete

Deleting an agent sets is_active=false. Inactive agents return HTTP 410 on chat and protected endpoints.

Fund before writes

Transact tools check wallet balance and spending caps before broadcast. Fund the agent on each target chain before asking it to send tokens or contract calls.