Skip to content

Architecture

This section describes the internal architecture of Lango, a Go-based AI agent built on Google ADK v1.0.0

  • System Overview


    High-level architecture layers, component interactions, and a visual diagram of how Lango processes messages from channels through the agent runtime to AI providers.

  • Project Structure


    Complete directory layout with descriptions of every package, explaining what each module owns and how packages relate to each other.

  • Data Flow


    End-to-end message flow, the bootstrap/wiring process, and async buffer patterns used for embedding generation and graph indexing.

  • Master Document


    Top-level product constitution, product path, capability areas, and execution-track portfolio for Lango.

  • External Collaboration Audit


    The first audit ledger for the product area that most directly defines Lango: trust, pricing, settlement, teams, and shared artifacts.

  • Trust, Security & Policy Audit


    Detailed audit ledger for the policy and safety boundaries that determine whether early knowledge exchange is actually safe to operate.

  • Identity Trust Reputation Audit


    Audit ledger for identity continuity, trust entry, reputation, and revocation in knowledge exchange v1.

  • Pricing Negotiation Settlement Audit


    Audit ledger for pricing surfaces, negotiation, settlement, and escrow in knowledge exchange v1.

  • Knowledge Exchange Runtime


    The first transaction-oriented runtime control plane for knowledge exchange v1, centered on transaction receipt and submission receipt with explicit current limits.

  • :material-bank-check-outline: Settlement Progression


    The first transaction-level settlement progression slice for knowledge exchange v1, covering approve, revise, reject, and escalate with explicit current implementation limits.

  • Actual Settlement Execution


    The first direct settlement execution slice for knowledge exchange v1, connecting approved-for-settlement state to real payment execution with explicit current limits.

  • Partial Settlement Execution


    The first direct partial settlement execution slice for knowledge exchange v1, executing one canonical partial amount with explicit current limits.

  • Escrow Release


    The first funded-escrow release slice for knowledge exchange v1, connecting funded escrow and approved settlement state to real release execution with explicit current limits.

  • Escrow Refund


    The first funded-escrow refund slice for knowledge exchange v1, connecting review-path funded escrow to refund execution with explicit current limits.

  • Dispute Hold


    The first dispute-linked escrow hold slice for knowledge exchange v1, recording hold evidence for funded dispute-ready escrow with explicit current limits.

  • Release vs Refund Adjudication


    The first post-hold adjudication slice for knowledge exchange v1, recording canonical release-vs-refund branching without yet executing either path.

  • Adjudication-Aware Release/Refund Execution Gating


    The first slice that connects canonical escrow adjudication to the existing release and refund executors with explicit branch gating and one-way safety rules.

  • Automatic Post-Adjudication Execution


    The first inline convenience slice that lets escrow adjudication optionally execute the selected release or refund path immediately while preserving existing executor gates.

  • Background Post-Adjudication Execution


    The first async convenience slice that lets escrow adjudication enqueue the selected release or refund path onto the existing background task substrate.

  • :material-refresh-alert: Retry / Dead-Letter Handling


    The first retry slice for background post-adjudication execution, adding bounded exponential backoff and terminal dead-letter failure without changing canonical adjudication.

  • Operator Replay / Manual Retry


    The first operator-facing replay slice for dead-lettered post-adjudication execution, reusing canonical adjudication and the existing background dispatch path.

  • Policy-Driven Replay Controls


    The first authorization slice for replay, adding actor- and outcome-aware allowlists on top of the canonical replay gate.

  • Dead-Letter Browsing / Status Observation


    The first read-only operator visibility slice for actor/time-aware dead-letter backlog browsing and per-transaction post-adjudication execution status.

  • P2P Knowledge Exchange Track


    The first concrete product track for external sovereign-agent economic activity.