What's new in AI Agentics
Every meaningful change to the AI Agentics platform, in one place. New features, improvements, fixes, and security updates — shipped continuously and documented here so you always know what changed and why it matters.
- Updated Jun 2026
- Ships continuously
- SOC 2 Type II
Welcome to the AI Agentics changelog — the canonical record of every product update we ship. We release small and often, so rather than batch changes into quarterly notes, we log them here as they go live. Whether you build single agents or coordinate multi-agent systems, this is the fastest way to see what changed across the platform, SDKs, and integrations.
Entries are tagged so you can scan at a glance: New for capabilities that didn't exist before, Improved for upgrades to existing features, Fixed for bug fixes and reliability work, and Security for compliance and safety updates. Want a look further ahead? The roadmap covers what's coming next.
2026 product updates at a glance
A snapshot of what we've shipped across the platform in the first half of the year.
Major releases
Jan–Jun 2026
New integrations
tools & connectors
Lower inference cost
via model routing
Type II certified
zero exceptions
Multi-agent teams
Typed handoffs, scoped memory, and per-agent budgets make orchestrating specialist agents safe and predictable.
Observability
End-to-end tracing of every reason → act → observe step, with cost, latency, and one-click run replay.
Trust & safety
SOC 2 Type II, PII redaction, tool allow-lists, and human-in-the-loop gates on high-risk actions.
AI Agentics changelog — full release log
Newest first. Each entry links back to deeper detail in the docs where relevant.
- Jun 2026New
Multi-agent teams v2 with typed handoffs
Compose orchestrator-and-worker teams declaratively. Agents now pass typed payloads on handoff, share a scoped memory namespace, and surface per-agent token and latency budgets so a single runaway worker can't blow the run.
- May 2026New
Smart model routing across providers
A new routing layer picks the cheapest model that clears a per-step quality bar, with automatic fallback on rate limits or timeouts. Early adopters cut inference spend 38% while holding eval pass rates within 1 point.
- May 2026Improved
Distributed tracing for every tool call
Traces now capture the full reason → act → observe loop, including nested sub-agent spans, retried tool calls, and the exact prompt and token counts at each step. Filter by status, cost, or duration and replay any run from a span.
- Apr 2026New
Eval suite with LLM-as-judge and regression gates
Build datasets from production traces, score agents with rubric-based LLM judges or exact-match checks, and block deploys that regress on a saved baseline. CI integration ships with the GitHub Action and CLI.
- Apr 2026Improved
20 new tool integrations and an OpenAPI importer
Added native connectors for Postgres, Snowflake, Linear, Notion, Stripe, and more. Paste any OpenAPI spec to generate typed tools automatically, complete with auth scopes and rate-limit handling.
- Mar 2026New
Managed vector memory for long-running agents
A built-in vector store gives agents persistent long-term memory with semantic recall, automatic chunking, and per-tenant isolation. Wire RAG into any agent without standing up your own embeddings pipeline.
- Mar 2026Improved
TypeScript and Python SDK 2.0
Both SDKs ship streaming-first APIs, strongly typed tool definitions, structured error types, and first-class support for the agent loop. Migration codemods and a side-by-side upgrade guide are in the docs.
- Feb 2026New
Template gallery for common agent patterns
Launch a support triage, research, or coding agent in one click. Templates bundle the prompt, tools, eval set, and guardrails as editable code so teams start from a working baseline instead of a blank file.
- Jan 2026New
SOC 2 Type II and granular guardrails
Completed our SOC 2 Type II audit with zero exceptions. Shipped configurable guardrails: PII redaction, tool allow-lists, output schema validation, and human-in-the-loop approval gates on high-risk actions.
- Jan 2026Fixed
Stable streaming under concurrent load
Resolved dropped tokens and premature stream closes that surfaced when a single agent fanned out many parallel tool calls. Streaming connections now back-pressure cleanly and reconnect without losing partial output.
Never miss a release
We ship updates every week. Follow the public roadmap to see what's in flight and vote on what we build next, or watch the documentation for migration guides as new SDK versions land. Building production agents? The API reference tracks every endpoint change alongside this log.
Make the most of every update
Roadmap
See what's planned next and shape priorities by voting on the features that matter to your team.
Learn moreDocumentation
Guides, migration notes, and references kept in lockstep with every release shipped here.
Learn moreTemplates
Start from a working agent — support triage, research, or coding — instead of a blank file.
Learn moreBuild on the latest AI Agentics release
Every feature in this changelog is live and ready to use. Start building production-grade agents today — free to begin, no credit card required.