SDKs in four languages
Service records, signed transaction context, durable delivery and client capture for .NET, Java, Python and Node.js.
Integration
Connect your applications and institutional systems
Collect records through .NET, Java, Python and Node.js SDKs, the OTLP/HTTP JSON collector and the PostgreSQL source reader. Use defined integration paths for HTTP, messaging, AI frameworks and institutional systems.
ARCHITECTURE VIEW / 03
Integrate the shared service once
Gateways, AI services and business services used by multiple projects share an integration. A new project joins through its entry point, transaction context and recording scope.
SHARED SERVICE INTEGRATION WITHIN ONE INSTITUTION AND ENVIRONMENT
PROJECT A
Digital channel
Entry point + transaction context
PROJECT B
Decision application
Entry point + transaction context
NEW PROJECT
New application
Entry point + recording scope
SHARED LAYER
API gateway
One integration setup
Application identity and context propagation
SHARED SERVICES
AI and business services
SDK / binder per service
Reused across projects
SIGILLUM FİNANS
Transaction review by application
01
Shared service setup
The capture method, service connection and route configuration are established on the shared service.
02
The new project’s context
The application entry point, context-carrying steps and expected records are configured for the project.
03
One institution portal
Authorized institution users select an application and review shared-service records within the corresponding transaction flow.
INTEGRATION SUPPORT
Supported technologies
| TECHNOLOGY | VERSION / METHOD | CAPABILITY |
|---|---|---|
| .NET | .NET 8 / 9 / 10 · IHttpClientFactory | Record HTTP requests and responses, with redirect and retry steps linked to the transaction. |
| Java | Java 8 / 11 / 17 / 21 / 25 · JDK HTTP · HC5 · OkHttp · WebClient | Core SDK, HC5 and OkHttp on Java 8; JDK HTTP on 11; Spring WebClient on 17. |
| Python | Python 3.9–3.14 · HTTPX · requests · aiohttp | Capture HTTP exchanges and deliver records through a durable local queue. |
| Node.js | Node.js 20 / 22 / 24 · fetch · undici · axios | Capture HTTP exchanges at the client transport layer and link individual attempts. |
| RabbitMQ | .NET · Java · Python · Node.js | Connect publish, delivery, acknowledgement and retry records, including outbox and duplicate-delivery handling. |
| Kafka | .NET · Java · Python · Node.js | Track messages by topic, partition, offset and consumer group, with explicit commit records. |
| IBM MQ | .NET · Java (javax / Jakarta JMS) | Adapters cover publish, commit, backout and backout-queue observation. Live queue-manager acceptance is a deployment step. |
| OpenTelemetry | OTLP / HTTP JSON · GenAI | The collector maps GenAI spans to model, retrieval and tool records using a published field mapping. |
| AI frameworks | LangChain (Python) · Microsoft.Extensions.AI (.NET) | Framework callbacks record model interactions; LangChain also records retrieval and tool calls. Each record identifies its observer. |
| Model API formats | OpenAI Chat Completions / Responses · Anthropic Messages | The installation derives model records from completed, sealed HTTP exchanges on configured routes. |
| Corporate identity | OpenID Connect · SAML 2.0 | Connect the institution’s identity provider to portal sign-in, with role mapping and MFA policies. |
| Security monitoring | JSON · NDJSON · CEF · Syslog | Share signed audit and alert feeds with SIEM systems through scoped APIs and configured syslog delivery. |
| CMDB · GRC · ITSM | Standard API · JSON / CSV | Provide inventory and control data to institutional systems; link external asset and incident identifiers. |
| Model, document and rule registries | Standard HTTPS contract · Local directory | Seal the response for a selected version and compare it with later registry checks. |
| Institutional archive | S3 API · Object Lock | Deliver evidence to the institution’s archive and retain object-version, retention and retrieval verification records. |
Integration planning matches these methods to your institution’s versions, client configuration and recording requirements. Deployment includes acceptance in your own environment.
01 / Record collection
SDKs, collector and source reader
Integration starts by defining transaction entry points and the expected services, events and fields. Applications use shared recording contracts to send transaction context and supporting information.
Shared services are integrated once within the same institution and environment. New applications reuse those connections through their entry points and recording scopes, with their own business steps added to the scope.
OpenTelemetry collector
GenAI traces received through JSON-encoded OTLP/HTTP are mapped to model, retrieval and tool-call records.
PostgreSQL source reader
Source inventories from a read-only table or view, delivered for reconciliation with a receipt from a separate source witness.
02 / HTTP capture
Capture at the supported client integration point
HTTP capture links requests and responses from supported clients to transaction context. Client-specific integration points correlate records of redirect and retry steps.
Installation reports show the capture method, record counts and detected scope conditions. Teams can assess the integration against actual application workflows.
.NET
HttpClient configured through IHttpClientFactory, with incoming request integration for ASP.NET Core.
Python
Explicit or automatic capture for httpx, requests and aiohttp.
Node.js
Client-specific capture integration for fetch, axios and undici.
Java
JDK HttpClient, the Apache HttpClient 5 classic API, OkHttp and Spring WebClient with the defined Reactor Netty configuration.
03 / Messaging
Publishing, delivery and processing records
Messaging adapters link publishing, broker responses, consumer delivery and processing outcomes to the same transaction. Retries, outbox relay and duplicate deliveries identified by the application’s inbox are tracked as distinct records.
The outbox model retains the business service as the owner of the message and records relay sends as attempts of the same publication. Consumer acknowledgements and processing outcomes have their own fields.
RabbitMQ
Publishers and consumers for .NET, Java, Python and Node.js, with pull or subscription delivery, explicit acknowledgement and dead-letter observation.
Kafka
Publishers and consumers for .NET, Java, Python and Node.js, recording topic, partition, offset, consumer group and explicit offset acknowledgement.
IBM MQ
.NET and Java javax/Jakarta JMS adapters for queues and topics, commit/backout and backout-queue observation. Real queue-manager acceptance is completed during institutional integration.
04 / AI frameworks
LangChain and Microsoft.Extensions.AI
The Python LangChain binder connects model calls, retrieval results and tool usage to the recording flow. The .NET Microsoft.Extensions.AI binder records IChatClient conversations with their transaction context.
Application-reported prompt template names and versions, and documents’ own identifiers and versions, are added to the record. Framework observations and records produced from sealed HTTP exchanges retain their respective provenance.
05 / Institutional registries
Model, document and rule references
The registry contract queries a named model, document or rule version in the institution’s record system. Connections use a configured HTTPS endpoint or an offline directory export prepared by the institution.
The response identifier, version and content hash are retained in a sealed reference. Historical queries compare new and earlier responses; optional content measurement records the source-reported hash and the measured hash separately.
06 / Institutional systems
CMDB, GRC, SIEM and ITSM connections
Standard APIs expose asset inventories, control statements, alerts and audit records to institutional systems. External system identifiers can be linked to Sigillum objects, while incident numbers and links can be added to alert tracking.
Signed HTTPS webhooks, SMTP and Syslog/CEF channels support notifications and record transfer. Machine access uses scoped integration credentials, and connections are configured for the institution’s internal services and workflows.
PRODUCT INFORMATION
Frequently asked questions
Which runtimes do the SDKs support?
.NET libraries target net8.0 and net10.0 and run on .NET 8, 9 and 10. The Java core targets 8, the JDK HttpClient module 11 and the Spring WebClient module 17. Python starts at 3.9 and Node.js at 20. Client-library and module versions are selected together using the support matrix.
How can an existing OpenTelemetry setup be used?
The collector maps GenAI fields sent through JSON-encoded OTLP/HTTP to model, retrieval and tool-call records. Service identity, transaction context and field mappings are configured, and records reach the central service through durable delivery.
What work does a new application require?
Shared service integrations are reused within the same institution and environment. Define the application’s entry points, service relationships and recording scope, then add its own business steps and new service connections to that scope.
How is integration acceptance performed?
After selecting supported clients and versions, teams configure recording scope, source mappings and institutional connections. Record correlation, scope checks and error behavior are assessed with real application traffic, while performance and capacity are measured in the target environment.