October 2026 Summaries

Apollo has introduced the GraphOS Router Support Tool, which creates on-demand, sanitized diagnostic bundles for Kubernetes-hosted GraphOS Router deployments without disrupting router processes. Built on the open-source troubleshoot.sh framework, the tool gathers router versions and configurations, current and prior container logs, pod health and resource data, optional Prometheus metrics, and sanitized router.yaml files into a single archive stored locally or in customer-controlled object storage. Automatic redaction removes generic secrets, credentials, connection strings, JWT and authentication settings, header values, GraphQL operation bodies, subgraph URLs, TLS keys, and Redis credentials before bundles are written. Users can run collection locally through kubectl or as an in-cluster Kubernetes Job, with both methods collecting the same data while supporting different access and storage requirements. Apollo receives bundles only if customers choose to share them, and the initial release does not support non-Kubernetes deployments or the Apollo GraphOS Operator, though operator support is planned.

Apollo has introduced GraphOS Factory, a preview agent skill that helps engineers convert existing REST APIs into Apollo Connectors subgraphs through an iterative, Git-backed workflow. Using an OpenAPI or Swagger specification or public API documentation, it inventories API operations and guides users through domain-modeling decisions such as scope, field meanings, relationships, mutations, PII handling, batching, and pagination rather than simply generating a wrapper around endpoints. The skill records decisions, pinned source materials, and discovered API behavior in a dedicated Git workspace, allowing work to be reviewed, resumed, shared across teammates and agent tools, and updated as APIs or Connector capabilities change. It validates generated subgraphs through composition, Connector unit tests, mocked Router tests, and optionally live API tests before exporting a subgraph ready for a supergraph. Available through Apollo Skills, GraphOS Factory has been evaluated on AppWorld and used internally for more than 40 services, though Apollo describes it as early-stage and is seeking feedback on its iterative approach.

Apollo’s GraphOS team describes how its Runtime Testing Framework evaluates whether new router releases produce genuine performance changes rather than ordinary benchmark noise, using isolated Kubernetes environments, deterministic mocked subgraphs, separate load-generator nodes, and five repeated runs per configuration. For version 2.18.0, it tested 10 graphs across five router configurations, assessing latency distributions, router p99, client latency, memory use, and HTTP and GraphQL error rates through three sequential gates: validating that each run reached stable target load without resource constraints, confirming that repeated measurements agreed, and determining whether observed changes exceeded predefined meaningful-change zones. Of 500 runs, 404 fully passed initial validity checks, while resource limitations and insufficient warm-up periods exposed improvements needed in the testing environment; repeatability checks also showed that metrics such as p99 required more samples to judge reliably. The analysis found no regressions in any comparable metric, with pooled median and p95 client latency within 2% of version 2.17.0, although many individual comparisons remained inconclusive rather than being treated as evidence of no change. The team subsequently strengthened its methodology with longer warmups, finer latency histograms, stricter thresholds, and plans for historical cross-release comparisons, reduced infrastructure noise, and guidance for customers tuning routers under load.

GraphOS Router 3.0, released in Preview on October 7, 2026, introduces deployment, performance, resilience, and GraphQL-specification updates for teams operating federated GraphQL APIs. Expanded Graph Artifacts can now package persisted queries with a supergraph schema, allowing routers to start from artifacts delivered through an organization’s own deployment infrastructure without contacting Apollo’s management plane. An incremental query planner aims to produce more efficient execution plans with substantially lower memory use and faster processing for large queries, while a rewritten Rust/Tower request pipeline and improved JSON fragment handling reduce buffering, coordination, and CPU work. Planned native circuit breaking for subgraphs and Apollo Connectors will isolate failing downstream services, and Redis-backed distributed rate limiting will enforce shared limits across router fleets, clients, tenants, incoming operations, and downstream calls, with a proposed observation mode for testing policies. Router 3.0 also adopts recent GraphQL specification features, including validation of the @oneOf directive, clearer @deprecated behavior, operation descriptions, and expanded Unicode support, with migration guidance available for Router 2.x users.

Apollo Connectors v0.4 is now generally available, allowing subgraph schemas to expose REST and HTTP APIs as typed GraphQL without resolver code or standalone subgraph servers. The release adds virtual connectors that generate data without HTTP requests, expanded inline transformations, a unified JSON- and GraphQL-inspired mapping grammar, and support for GraphQL interfaces and unions when mapping polymorphic REST responses. Enhanced static analysis identifies the output shapes and input dependencies of mappings during composition, improving diagnostics for invalid mappings and enabling type-aware transformation chains. The changes are intended to simplify connector authoring for both developers and AI agents, while allowing the same API layer to serve GraphQL clients and tools through Apollo MCP Server. Version 0.4 changes primitive literals in value positions to be interpreted as literals rather than field lookups, but Apollo reports that 97.4% of more than 25,000 analyzed selections behave identically across versions; the connect-migrate tool can identify and update affected schemas. Existing v0.2 and v0.3 schemas remain supported, while a future v0.5 release for Router v3 is planned to add developer-facing type warnings, reusable selection definitions, and additional transformation capabilities.

Apollo announced the public preview of GraphOS Agent Services, a platform intended to help organizations govern AI agents’ access to internal data and tools as deployments expand beyond simple prototypes. The service uses a graph-based domain model to connect enterprise systems, entities, and relationships, enabling agents to search for approved tools and data while applying identity-aware, field-level policies rather than granting or blocking access to entire API responses. Its capabilities include brokered credentials for agents and users, classifications and policies that can mask, deny, or allow sensitive fields, centralized suspension of agent identities, access-request workflows, and audit records showing requests, policy decisions, systems accessed, and withheld data. Apollo argues that these controls can make security reviews more manageable by preventing unauthorized data from entering an agent’s context and providing traceability for compliance and investigations. Intuit is cited as an early production GraphOS customer and preview participant, using the services to support real-time marketing analysis while limiting agent access and retaining audit trails.

Apollo has introduced its Runtime Testing Framework (RTF), a compositional, declarative toolkit for testing deployed services in production-like distributed environments, initially supporting GraphOS Router performance and release validation. Created to replace more fixed, manually processed router test setups, RTF allows engineers to define reproducible Test Plans containing an Environment, typically specified with Docker Compose or Kubernetes manifests, and a Scenario that runs test commands in containers. The framework supports reusable file providers, variable templating, and matrix configurations to parameterize tests across supergraphs, resource limits, traffic patterns, and deployment settings. Engineers can validate plans locally through a CLI using Docker Compose, then run them at scale through an internal Orchestrator that expands test variants and deploys each into an ephemeral Kubernetes namespace using Argo Workflows. Results, logs, outputs, and metrics are collected and stored for review through an internal interface, while Kubernetes and related technologies handle infrastructure, secrets, deployment, and metrics collection. Apollo plans to describe its specific use of RTF for GraphOS Router performance testing in a follow-up post.

Apollo’s Summit 2026 announcements center on extending its GraphOS platform to provide AI agents with governed, secure, and auditable access to enterprise APIs, addressing risks such as sensitive-data exposure, unclear API relationships, and lost user identity. GraphOS Agent Services, now in public preview, uses a domain model and adds search, identity, policy, and audit capabilities to control agent access at the field level, broker credentials, preserve user identity, and record activity. Apollo also previewed GraphOS Desktop, an agent-assisted environment for building domain models from existing APIs, while GraphOS Factory and expanded MCP Server tooling help teams convert REST APIs into graph components and manage GraphOS through natural language. Other updates include GraphOS Router 3.0, which introduces circuit breaking, a faster incremental query planner, Federation 3 support, and improved artifact delivery; generally available variant API keys for tighter environment isolation; a diagnostic Router Support Tool; and a Runtime Testing Framework for release validation. Developer-focused releases include Apollo Connectors v0.4, Apollo Client 4.3 custom scalars, experimental Apollo Mock for Kotlin, Rover 1.0 with OAuth authentication, improved schema proposal reviews, and Kubernetes deployment support for Apollo MCP Server through GraphOS Operator.

Apollo has introduced an opt-in Incremental Query Planner in GraphOS Router 3.0 to address long planning times and high memory consumption for complex federated GraphQL graphs. Replacing the prior breadth-first approach, the new planner uses depth-first search to produce an initial plan quickly, mutates a single in-memory Fetch Graph rather than repeatedly copying candidate plans, and applies BULB search to bound memory while retaining backtracking opportunities for better plans. Configurable fuel and beam-width settings let users balance planning quality, computation time, and memory use, with fuel limiting how extensively the planner explores alternatives after its initial draft. In tests across 200,000 operations from highly complex customer graphs, Apollo reports that the planner completed even previously minute-long operations in under two seconds, delivering up to a 300-fold speedup and 97% lower memory usage, though some large operations may require higher fuel settings to find optimal plans. The feature includes metrics for tuning and is available for feedback before potentially becoming the default planner.

Apollo has released Rover 1.0, a major update to its command-line tool for managing Apollo Federation and GraphOS workflows such as publishing subgraphs, running schema checks, and composing supergraphs. The release introduces OAuth-based interactive authentication alongside improved API key and grant management, while replacing scattered environment-variable configuration with inspectable profiles for graphs and environments. Rover 1.0 also removes support for Apollo Federation 1, encouraging remaining users to migrate to Federation 2, and strengthens its plugin system through pinned versions, reproducible behavior, and transparent downloads. New native support for contract variant results in subgraph checks and publishing helps teams assess changes across consumer-specific graph views, while client operation checks have moved from the now-deprecated Apollo CLI into Rover. Because the version removes Federation 1 support and changes configuration practices, Apollo recommends reviewing its migration guide before upgrading, particularly for CI environments.