Sageo
Trust and discovery infrastructure for AI agents on MOI Chain.
- Status
- Collaboration · MOI Labs × BoilerBlockchain
- Role
- Project Lead / SDK Engineer
- Timeline
- Sep–Dec 2025
- Stack
- TypeScript · Node.js · MOI Chain · Google A2A Protocol · Coco · React
01Context
Sageo addresses a gap in the emerging A2A agent communication ecosystem: agents can talk to each other, but there is no standard way to verify who they are, what they have done, or whether a claimed interaction actually occurred.
Sageo adds that layer — a registry for agent identity and a tamper-evident log of interaction proofs, backed by MOI Coco smart contracts and surfaced through a TypeScript SDK, an Express REST API, and a React discovery UI.
02On-chain identity and interaction proofs
Two Coco contracts form the core. SageoIdentityLogic registers agent metadata — AgentCard-style profiles, skills, URLs, and ownership. SageoInteractionLogic stores per-agent interaction records keyed by actor.
The interaction records use SHA-256 payload hashes with A2A context, task, and message identifiers rather than raw request and response bodies. Hashing reduces the on-chain payload size; it does not by itself establish payload confidentiality or a measured transaction-cost saving.
03TypeScript SDK and A2A wrappers
The SDK exposes two wrapping primitives: an outgoing A2A client wrapper and a server-side request handler wrapper. Both inject a Sageo trace metadata envelope under a well-known extension key, hash the request and response payloads, and log the proofs to the interaction contract.
The outgoing wrapper resolves remote Sageo IDs by agent URL when needed, logs before the A2A call, and logs response or error after. The server-side wrapper reads Sageo metadata from incoming A2A messages and logs the inbound interaction symmetrically.
SDK design favours availability: if Sageo logging fails, the wrapper warns and continues rather than blocking the underlying A2A call.
04Explorer API and discovery UI
An Express REST API exposes agent discovery, search, profile and card lookup, skill filtering, interaction history, and stats — all backed by MOI contract reads. Off-chain filtering and pagination are handled in the API layer rather than in contracts, an intentional design decision documented in the spec to keep contract logic minimal.
A React/Vite frontend provides discovery with search and tag filtering, agent profile views with success rate and network reach stats, and a registration flow that returns one-time wallet credentials.
05Devnet edge cases and fallbacks
MOI devnet surfaced receipt decoding inconsistencies during development. The SDK and API include defensive fallbacks: multiple receipt extraction paths, POLO-output decoding fallbacks, and ephemeral-state fallback reads when actor storage is not yet initialised.
The repository includes E2E workflow scripts intended to check a mock A2A interaction against stored hashes, caller/callee identifiers, and interaction counts. Those scripts describe a configured devnet verification path, rather than proving that every deployment has passed it.
06Known limitations
Devnet only. Logic IDs and mnemonics are hardcoded for MOI devnet. A production deployment requires env-managed keys and mainnet contract addresses.
In-memory discovery filtering. The API fetches agent IDs and filters and paginates results in memory. Indexed queries or caching would avoid repeating that work as the registry grows; the repository does not establish a measured capacity threshold.
Registration credential handling. The registration API returns a generated wallet mnemonic to the client. Production use would require a deliberate credential-provisioning and custody design; the prototype flow is not a security guarantee.