A2A Daemon Engine

Package Version: 0.0.1 A2A SDK: v1.0.2 (a2a-sdk[http-server]) Status: Phases 1–9 complete; Phase 10 (gateway-mediated ai_agent_core_engine integration) planned Last Updated: 2026-07-14

An Agent-to-Agent (A2A) protocol daemon engine module for the SilvaEngine Gateway (silvaengine_gateway). It is not a standalone service — it is loaded as a registered gateway module and all client access flows through the gateway's HTTP, WebSocket, and SSE routes. The gateway handles authentication, routing, SSE client management, and dispatch; this module provides the A2A protocol logic, agent execution, task persistence, and LLM handler bridging.

Built on the official A2A SDK server pattern, with dual-backend persistence (DynamoDB + PostgreSQL), multi-tenant isolation, and pluggable LLM handler integration.

How It Integrates with the Gateway

The daemon is registered as a SilvaEngine Gateway module via deploy() in main.py, which declares three gateway-facing entry points:

Entry Point	Gateway Route	Method	Purpose
`a2a_core_graphql`	`POST /{ep}/a2a_core_graphql`	POST	GraphQL CRUD for agents, tasks, messages, settings
`a2a`	`POST /{ep}/a2a`	POST	A2A JSON-RPC protocol (`message/send`, `message/stream`, `tasks/get`, `tasks/cancel`)
`sse_message`	`POST /{ep}/sse`	POST	A2A JSON-RPC + push activity to SSE clients (partition-scoped)

The gateway additionally exposes:

Gateway Route	Method	Purpose
`GET /{ep}/sse`	GET	SSE stream — gateway-managed `sse_manager` delivers live A2A task events to connected clients
`GET /{ep}/a2a_sse`	GET	A2A-specific SSE stream (alias)

The daemon does not listen on its own port or run its own HTTP server in production. All transport, auth, and SSE client lifecycle is owned by silvaengine_gateway. The daemon's internal SDK Starlette app and FastAPI operations app exist for local development and testing only.

Architecture — Gateway Dispatch Flow

A2A Client
    │
    ▼ POST /{ep}/a2a  (message/send | message/stream)
silvaengine_gateway  (auth, routing, SSE client mgmt)
    │
    ▼ dispatch → A2ADaemonEngine.a2a(**params)
A2ADaemonExecutor
    │
    ├── Non-streaming → handler.ask_model(input_messages, context)
    │       └── GraphQL → ai_agent_core_engine → final_output
    │
    └── Streaming → handler.ask_model(..., stream_queue, stream_event)
            └── WebSocket → ai_agent_core_engine → chunk_delta frames
                    └── Dual-path: SDK EventQueue + sse_manager.broadcast_to_partition()

Client receives:
    • Non-streaming: single JSON-RPC response with A2A Message
    • Streaming: live SSE events via GET /{ep}/sse (gateway-managed)

Local Development Standalone Mode

For local development and testing, the daemon can run as a standalone HTTP server using the internal SDK Starlette app:

Route	Method	Purpose
`/.well-known/agent-card.json`	GET	Public Agent Card (protocol version `1.0.0`)
`/`	POST	JSON-RPC compatibility endpoint (slash-style: `message/send`, `tasks/get`, `tasks/cancel`)
`/v1`	POST	SDK native JSON-RPC dispatcher (`SendMessage`, `GetTask`, `CancelTask`)
`/tasks/{task_id}/stream`	GET	SSE task streaming with `Last-Event-ID` replay buffer
`/rest/health`	GET	Health check
`/rest/{endpoint_id}/a2a_core_graphql`	POST	GraphQL API for agents, tasks, messages, settings

This mode is for development only — production deployments go through the gateway.

Key Features

A2A SDK v1.0 — JSON-RPC over HTTP (slash-style at /, native at /v1)
Gateway module — registered via deploy(), dispatched by silvaengine_gateway
Public Agent Card at /.well-known/agent-card.json with ETag / Last-Modified
SDK AgentExecutor integration (A2ADaemonExecutor)
Dual-backend persistence — DynamoDB (PynamoDB) or PostgreSQL (SQLAlchemy + Alembic)
Multi-tenant isolation via composite partition keys ({endpoint_id}#{part_id})
SSE streaming — gateway-managed sse_manager delivers live tokens to connected clients
GraphQL operations API for CRUD on agents, tasks, messages, settings
Dual authentication — local JWT (HS256) and AWS Cognito (RS256 + JWKS)
Pluggable LLM handlers — per-agent module_name / class_name selection
Hermes Agent bridge — route A2A tasks to Hermes Agent API Server via HTTP + SSE
Gateway-mediated core engine bridge (Phase 10, planned) — GraphQL for non-streaming, WebSocket for streaming
Production hardening — rate limiting, health monitoring, circuit breakers, OpenTelemetry
Experimental gRPC transport — JSON-over-gRPC with bidirectional streaming
AWS Lambda support — serverless JSON-RPC dispatch via A2ADaemonEngine.a2a(**event)

Project Structure

a2a_daemon_engine/
├── a2a_daemon_engine/
│   ├── main.py                          # A2ADaemonEngine — gateway module entry, deploy(), serverless dispatch
│   ├── schema.py                        # GraphQL schema (Query, Mutations, types)
│   ├── handlers/
│   │   ├── config.py                    # Config singleton — backend, auth, Phase 10 settings
│   │   ├── a2a_server.py                # Builds SDK Starlette app + DefaultRequestHandler
│   │   ├── a2a_executor.py              # A2ADaemonExecutor(AgentExecutor) — request routing
│   │   ├── a2a_handlers.py              # Business handlers (delivery, retry, routing)
│   │   ├── a2a_jsonrpc_bridge.py         # JSON-RPC → SDK request normalization
│   │   ├── a2a_taskstore.py              # DynamoDBA2ATaskStore (SDK TaskStore)
│   │   ├── a2a_sse.py                   # SSEEventQueue, StreamingTaskManager, stream routes
│   │   ├── sse_manager.py               # Gateway-compatible SSE client manager
│   │   ├── a2a_core.py                  # GraphQL handler for agents/tasks/messages/settings
│   │   ├── a2a_utility.py              # DynamoDB query/mutation helpers
│   │   ├── a2a_ai_agent_utility.py      # Phase 10 bridge utility (agent resolution, streaming)
│   │   ├── a2a_hermes_handler.py        # Hermes Agent bridge plugin (HTTP + SSE)
│   │   ├── a2a_extended_card.py          # Extended agent card with auth gating
│   │   ├── a2a_telemetry.py             # OpenTelemetry instrumentation (optional)
│   │   ├── a2a_health_monitor.py        # Health monitoring + circuit breakers
│   │   ├── a2a_rate_limiter.py         # Token-bucket rate limiting
│   │   ├── a2a_cancellation.py          # Task cancellation propagation
│   │   ├── a2a_pushconfig.py           # Push notification config + SSRF allowlist
│   │   ├── a2a_cost_extension.py       # Cost/quota tracking (scaffold)
│   │   ├── a2a_secure_passport.py      # Secure passport (scaffold)
│   │   ├── a2a_grpc.py                 # gRPC transport (experimental)
│   │   ├── a2a_graphql_subscriptions.py # WebSocket GraphQL subscriptions
│   │   ├── middleware.py               # FlexJWTMiddleware (local + Cognito)
│   │   ├── auth_router.py              # OAuth2 /auth/token endpoint
│   │   ├── jwt_local.py                # Local JWT (HS256)
│   │   ├── jwt_cognito.py             # Cognito JWT (RS256 + JWKS)
│   │   └── schema.py                  # GraphQL schema wiring
│   ├── models/
│   │   ├── dynamodb/                   # PynamoDB models (agent, task, message, setting)
│   │   ├── postgresql/                 # SQLAlchemy ORM models + base
│   │   └── repositories/
│   │       ├── base.py                 # EntityRepository abstract base
│   │       ├── dispatch.py             # get_repo() — backend dispatch
│   │       ├── dynamodb/               # DynamoDB repository implementations
│   │       └── postgresql/             # PostgreSQL repository implementations
│   ├── mutations/                      # GraphQL mutations (agent, task, message, setting)
│   ├── queries/                        # GraphQL queries (agent, task, message, setting)
│   ├── types/                          # GraphQL types (agent, task, message, setting)
│   ├── utils/                          # Exceptions, normalization helpers
│   └── tests/                          # Test suite (see Testing below)
├── migration/
│   ├── alembic.ini                     # Alembic config (PostgreSQL)
│   └── alembic/versions/               # 4 migrations (agents, tasks, messages, settings)
├── docs/                               # Architecture, protocol, development plan, integration
├── AGENTS.md                           # Contributor reference
├── pyproject.toml                      # Poetry config, deps, ruff, pytest
└── README.md                           # This file

Installation

Prerequisites

Python 3.10–3.12
Poetry (or pip)
silvaengine_gateway (for production deployment)

Setup

# Clone the repository
git clone https://github.com/ideabosque/a2a_daemon_engine.git
cd a2a_daemon_engine

# Install with Poetry
poetry install

# Or install with pip (including dev deps)
pip install -e .[dev]

# Optional extras
pip install -e .[grpc]          # gRPC transport
pip install -e .[telemetry]    # OpenTelemetry
pip install -e .[postgresql]   # PostgreSQL backend (SQLAlchemy + psycopg2 + Alembic)
pip install -e .[all]           # All extras

PostgreSQL Backend Setup

# Set backend
export DB_BACKEND=postgresql

# Run migrations
alembic -c migration/alembic.ini upgrade head

Configuration

Configuration is injected by the gateway at module initialization time (Config.initialize(logger, **setting)). The following settings are recognized:

# Backend selection: "dynamodb" (default) or "postgresql"
export DB_BACKEND=dynamodb

# AWS (DynamoDB backend)
export REGION_NAME=us-east-1
export AWS_ACCESS_KEY_ID=your_key
export AWS_SECRET_ACCESS_KEY=your_secret

# PostgreSQL backend (when DB_BACKEND=postgresql)
# Configure connection in alembic.ini or via DATABASE_URL

# Authentication (used by local dev server; gateway handles auth in production)
export AUTH_PROVIDER=local
# Required for AUTH_PROVIDER=local. Must be ≥ 32 chars and not a default/weak
# value (e.g. CHANGEME, secret, password).
export JWT_SECRET_KEY=replace-me-with-a-strong-random-secret-of-32-or-more-chars

# Phase 10: ai_agent_core_engine bridge (planned — gateway-mediated)
export A2A_DEFAULT_AGENT_UUID=a2a-default-agent
export A2A_STREAM_TIMEOUT=120.0
export A2A_STREAMING_ENABLED=true

# Phase 10: Hermes Agent bridge
export HERMES_API_URL=http://localhost:8642
export HERMES_API_KEY=your-api-key
export HERMES_MODEL=hermes-agent
export HERMES_STREAM_TIMEOUT=300

Deployment

Production — As a Gateway Module

In production, the daemon is loaded by silvaengine_gateway as a registered module. The gateway owns the HTTP listener, WebSocket connections, SSE client lifecycle, authentication, and routing. The daemon's deploy() function declares its entry points (a2a, a2a_core_graphql, sse_message) and the gateway dispatches incoming requests to them.

To register the module with the gateway, add it to the gateway's module configuration and ensure the package is importable in the gateway's Python environment.

AWS Lambda (Serverless)

A2ADaemonEngine.a2a(**event) accepts JSON-RPC 2.0 dictionaries only — non-JSON-RPC payloads are rejected. The bridge in a2a_jsonrpc_bridge.py normalizes the request into SDK protobuf types and dispatches to DefaultRequestHandler.

from a2a_daemon_engine.main import A2ADaemonEngine

daemon = A2ADaemonEngine(logger, transport="lambda", port=0, ...)

def lambda_handler(event, context):
    # event must be a JSON-RPC 2.0 dictionary
    return daemon.a2a(**event)

Local Development

Run the Standalone Dev Server

For local development, the daemon can run its own HTTP server with the internal SDK Starlette app and FastAPI operations app:

# Via Poetry
poetry run python -m a2a_daemon_engine.main

# Or via the dev script
python a2a_daemon_engine/tests/start_daemon.py

The dev server listens on http://0.0.0.0:${PORT:-8001}.

Verify (Local Dev Server)

# Health check
curl http://localhost:8001/rest/health

# Agent Card
curl http://localhost:8001/.well-known/agent-card.json

# Compatibility endpoint (slash-style methods)
curl -X POST http://localhost:8001/ \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc": "2.0", "method": "message/send", "params": {"message": {"role": "user", "parts": [{"text": "hello"}]}}, "id": 1}'

# SDK native dispatcher
curl -X POST http://localhost:8001/v1 \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc": "2.0", "method": "SendMessage", "params": {"message": {"role": "user", "parts": [{"text": "hello"}]}}, "id": 1}'

# SSE task stream (replace TASK_ID with a real task identifier)
curl -N http://localhost:8001/tasks/TASK_ID/stream

Verify (Via Gateway)

When deployed behind silvaengine_gateway, the same operations are available through gateway routes (replace <gateway-host>:<port> and <endpoint_id>):

# A2A JSON-RPC
curl -X POST http://<gateway-host>:<port>/<endpoint_id>/a2a \
  -H "Content-Type: application/json" \
  -H "Part-ID: <tenant_id>" \
  -d '{"jsonrpc": "2.0", "method": "message/send", "params": {"message": {"role": "user", "parts": [{"text": "hello"}]}}, "id": 1}'

# GraphQL
curl -X POST http://<gateway-host>:<port>/<endpoint_id>/a2a_core_graphql \
  -H "Content-Type: application/json" \
  -H "Part-ID: <tenant_id>" \
  -d '{"query": "{ a2aAgentList { agent_id name } }"}'

# SSE stream
curl -N http://<gateway-host>:<port>/<endpoint_id>/sse

Authentication

In production, the gateway handles authentication before dispatching to the daemon. The daemon's own JWT middleware (FlexJWTMiddleware) is used only in local development / standalone mode.

Two providers are supported, selected at startup with AUTH_PROVIDER:

Provider	Algorithm	Required env vars
`local`	HS256	`JWT_SECRET_KEY` (≥ 32 chars, must not be a default/weak value); optional `LOCAL_USER_FILE`, `ADMIN_STATIC_TOKEN`
`cognito`	RS256 + JWKS	`COGNITO_USER_POOL_ID`, `COGNITO_APP_CLIENT_ID`, `COGNITO_APP_SECRET`, optional `COGNITO_JWKS_URL`

Local users are obtained via POST /rest/auth/token (OAuth2 password grant, form-encoded). Public protocol routes (/, /v1, /.well-known/..., /tasks/{task_id}/stream) are not gated by FlexJWTMiddleware.

Multi-Tenancy

All persistence uses a composite partition key:

partition_key = "{endpoint_id}#{part_id}"

endpoint_id is taken from URL paths like /{endpoint_id}/a2a_core_graphql; part_id is read from the Part-ID request header (or query parameter when that path component is absent). The gateway passes partition_key into the dispatch call, and all data access flows through this composite key, so cross-tenant access is hard-isolated at the row level.

Dual-Backend Persistence

The daemon supports two database backends, selected via DB_BACKEND:

Backend	Config	Models	Migrations
DynamoDB (default)	`DB_BACKEND=dynamodb`	`models/dynamodb/` (PynamoDB)	Auto-create on startup
PostgreSQL	`DB_BACKEND=postgresql`	`models/postgresql/` (SQLAlchemy)	Alembic (`migration/alembic/`)

Repository selection is dispatched at runtime via models/repositories/dispatch.py → get_repo(entity_type), which lazily initializes the appropriate repository based on Config.DB_BACKEND.

Pluggable LLM Handlers

Agent execution is data-driven via per-agent module_name and class_name in the agent registry. The executor loads the handler dynamically and calls ask_model() with the same contract regardless of backend.

Hermes Agent Handler

Routes A2A tasks to a running Hermes Agent API Server via HTTP + SSE:

module_name: "a2a_daemon_engine.handlers.a2a_hermes_handler"
class_name:  "HermesAgentHandler"

Per-agent metadata: hermes_api_url, hermes_api_key, hermes_model, hermes_timeout. See Hermes Integration Guide for full setup, A2A state mapping, and E2E test instructions.

Core Engine Handler (Phase 10 — Planned)

Gateway-mediated integration with ai_agent_core_engine:

Non-streaming (SendMessage) → GraphQL mutation to POST /{ep}/ai_agent_core_graphql
Streaming (SendStreamingMessage) → WebSocket to /{ep}/ai_agent_core_ws with chunk_delta frames
Client-facing streaming stays on SSE (GET /{ep}/sse — gateway-managed)

See Development Plan — Phase 10 for the full design.

Testing

Run Tests

# All unit + skipped-live tests
python -m pytest a2a_daemon_engine/tests -q

# Live API tests (requires running dev daemon)
# On bash:
A2A_RUN_LIVE_API_TESTS=1 python -m pytest a2a_daemon_engine/tests/test_api.py -q
# On PowerShell:
$env:A2A_RUN_LIVE_API_TESTS='1'; python -m pytest a2a_daemon_engine/tests/test_api.py -q

# With coverage
python -m pytest --cov=a2a_daemon_engine --cov-report=html

# By marker
python -m pytest -m unit
python -m pytest -m "graphql or cache"
python -m pytest -m integration

Test Files

File	Coverage
`test_phase6.py`	A2A SDK v1.0 upgrade (state migration, enums, pagination)
`test_phase8.py`	Production hardening, SSE infrastructure, security
`test_phase9.py`	Advanced extensions (gRPC, subscriptions, health, rate limit)
`test_phase10.py`	Phase 10 bridge (agent resolution, streaming, persistence)
`test_hermes_handler.py`	Hermes Agent handler (mocked HTTP via `httpx.MockTransport`)
`test_executor_unit.py`	Executor unit tests
`test_handlers_unit.py`	Business handler unit tests
`test_a2a_jsonrpc_bridge.py`	JSON-RPC bridge normalization
`test_jwt_validation.py`	JWT validation (local + Cognito)
`test_api.py`	Live API integration tests
`test_postgresql_backend.py`	PostgreSQL backend tests
`test_helpers.py`	Fixture support for live-API suites

Test Markers

unit, integration, slow, a2a, agent, task, message, setting, server, graphql, cache, performance

Runnable Scripts (not pytest)

a2a_tck_checker.py — A2A TCK compliance checker
a2a_rpc_verifier.py — JSON-RPC protocol verifier
validate_agent_card.py — Agent Card validator
register_hermes_agent.py — Sample Hermes agent registration

Lint and Format

# Ruff (line length 88, target py310)
python -m ruff check a2a_daemon_engine/

# Black
python -m black a2a_daemon_engine/

# Mypy (ignore_missing_imports = true)
python -m mypy a2a_daemon_engine/

Development Phases

Phase	Theme	Status
1–3	Core SDK alignment (AgentExecutor, TaskStore, async GraphQL)	✅ Complete
4	Server restructuring (SDK app primary, FastAPI at `/rest`)	✅ Complete
5	Event-driven message delivery (retry + status tracking)	✅ Complete
6	A2A SDK v1.0 upgrade (state migration, enums, pagination)	✅ Complete
7	Streaming and multi-turn (SSE, INPUT_REQUIRED, AUTH_REQUIRED)	✅ Complete
8	Production hardening (extended cards, telemetry, TCK, security)	✅ Complete
9	Advanced extensions (gRPC, subscriptions, health, rate limit)	✅ Complete
10	Gateway-mediated `ai_agent_core_engine` integration (GraphQL + WebSocket)	📋 Planned

See the Development Plan for full details.

Documentation

Document	Purpose
Architecture	HTTP surface table, runtime components, request-flow sequence diagram
Protocol Call Flow	Per-method call paths for `message/send`, `tasks/get`, `tasks/cancel`
Development Plan	Phases 1–10, implementation notes, release gates
Test Plan	Unit / live / release-gate coverage
Integration Test Plan	End-to-end / integration playbook
Integration Scenarios SOP	Integration scenarios standard operating procedure
Hermes Integration	Hermes Agent bridge setup, state mapping, E2E tests
Hermes Bridge Dev Plan	Hermes bridge development plan
Protocol Analysis	Protocol background and design suggestions
Documentation Index	Documentation navigation guide
AGENTS.md	Contributor reference (environment, key files, JSON-RPC rules)
Tests README	Test suite overview and coverage

License

MIT

Navigation Menu

Uh oh!

A2A Daemon Engine

How It Integrates with the Gateway

Architecture — Gateway Dispatch Flow

Local Development Standalone Mode

Key Features

Project Structure

Installation

Prerequisites

Setup

PostgreSQL Backend Setup

Configuration

Deployment

Production — As a Gateway Module

AWS Lambda (Serverless)

Local Development

Run the Standalone Dev Server

Verify (Local Dev Server)

Verify (Via Gateway)

Authentication

Multi-Tenancy

Dual-Backend Persistence

Pluggable LLM Handlers

Hermes Agent Handler

Core Engine Handler (Phase 10 — Planned)

Testing

Run Tests

Test Files

Test Markers

Runnable Scripts (not pytest)

Lint and Format

Development Phases

Documentation

License

Uh oh!

FilesExpand file tree

README.md

Latest commit

History

README.md

File metadata and controls

A2A Daemon Engine

How It Integrates with the Gateway

Architecture — Gateway Dispatch Flow

Local Development Standalone Mode

Key Features

Project Structure

Installation

Prerequisites

Setup

PostgreSQL Backend Setup

Configuration

Deployment

Production — As a Gateway Module

AWS Lambda (Serverless)

Local Development

Run the Standalone Dev Server

Verify (Local Dev Server)

Verify (Via Gateway)

Authentication

Multi-Tenancy

Dual-Backend Persistence

Pluggable LLM Handlers

Hermes Agent Handler

Core Engine Handler (Phase 10 — Planned)

Testing

Run Tests

Test Files

Test Markers

Runnable Scripts (not pytest)

Lint and Format

Development Phases

Documentation

License