Context Ontology Accelerator

Summary

Context Ontology Accelerator is an open-source semantic context layer for AWS that enables agents to retrieve and validate context by combining knowledge graphs, formal ontologies, and rule-based systems. It is run as a service, following a Scan → Model → Serve workflow where it connects data sources, induces ontologies, and serves context via SPARQL federation to AI agents. The tool is intended for practitioners working with agent context management, and its documentation positions it alongside other graph databases for data-layer operations. The repository structure suggests an active development cycle with clear separation between core infrastructure and API model generation.

An open-source, ontology-based semantic context accelerator that enables AI agents to make more accurate, consistent, and explainable decisions.

Command line

Values are placeholders taken from each option’s declared default. Nothing is executed here — the output shown is a recording of a run that already happened.

Release history

1. v0.2.2 Aug 27, 2026 · issue 009

   Adds an iterative NL→SQL agent, FK-aware ontology graph traversal, and @pk/@fk(...) constraint tags for Athena connectors.

   └──▷ GET THIS VERSION
   copy

   $ git clone --branch v0.2.2 https://github.com/aws/context-ontology-accelerator.git
   # already have the repo? check out this version:
   $ git checkout v0.2.2

   • ›Adds explore_graph as an opt-in Tier-2 agent tool in OntologyGraphTool for foreign-key-aware join discovery, walking the ontology graph up to k hops from a seed table to surface related tables that vector search alone would miss.
   • ›Adds OntologyGraphTool.expand_from as a non-agent entry point to the same FK graph walk, enabling flat NL→SQL pipelines (which have no agent loop) to benefit from the same join-discovery traversal.
   • ›Adds a self-correcting NL→SQL agent (sql_agent.py) that iteratively calls search_tables, get_table_schema, generate_sql, and execute_sql — SQL execution is gated on an opaque handle returned by generate_sql, preventing arbitrary SQL from running.
   • ›Adds @pk and @fk(...) constraint tags parseable inside Athena column comments, enabling primary- and foreign-key metadata to travel through the Athena federation protocol (which has no native key-constraint field) and produce the same PrimaryKey/ForeignKey records as the JDBC path.
2. v0.2.1 Aug 20, 2026 · issue 004

   v0.2.1 adds a CloudWatch dashboard for the scan/enrichment pipeline and pagination support to the list_metrics MCP tool.

   └──▷ GET THIS VERSION
   copy

   $ git clone --branch v0.2.1 https://github.com/aws/context-ontology-accelerator.git
   # already have the repo? check out this version:
   $ git checkout v0.2.1

   • ›Adds nextToken pagination parameter to the list_metrics MCP tool, alongside renamed parameters namespaceId and maxResults (max 1000).
   • ›Adds a CloudWatch dashboard for the structured scan and enrichment pipeline, sourcing custom metrics from sources Lambdas via EMF-stdout.
   └──▷ BREAKING ON UPGRADE

   • !The list_metrics MCP tool parameters namespace_id and max_results are renamed to namespaceId and maxResults; any caller passing these by keyword name will break.
3. v0.2.0 Aug 13, 2026 · issue 004

   v0.2.0 adds Redshift Serverless as an alternative query engine and new CloudWatch guardrail observability metrics.

   └──▷ GET THIS VERSION
   copy

   $ git clone --branch v0.2.0 https://github.com/aws/context-ontology-accelerator.git
   # already have the repo? check out this version:
   $ git checkout v0.2.0

   • ›Adds executionEngine=REDSHIFT + redshiftWorkgroup onboarding parameters to route Glue-backed sources through Amazon Redshift Serverless via the redshift-data API instead of Athena, with automatic Trino→Redshift SQL transpilation and awsdatacatalog."<db>"."<table>" reference rewriting.
   • ›Adds guardrail observability under the COA/Guardrails CloudWatch namespace, emitting GuardrailInvocations (Count), GuardrailBlocked (Count), and GuardrailLatency (Milliseconds) metrics with support for both transport="put" (PutMetricData, used by ECS Fargate kg-build/enrichment/ontology tasks) and Embedded Metric Format via CloudWatch Logs.
   • ›Adds SQLFirewall.validate enforcement and dialect-aware LIMIT injection to the Redshift execution path, restricting queries to SELECT-only and preventing large scans from fully materializing.
   • ›Introduces CatalogConstraint and parse_referred_column imports alongside deterministic IRI minting via table_identity and _name_discriminator (SHA-256-derived 8-char suffix) for stable cross-run ontology graph generation.