The Anthropic Python SDK shipped v0.115.0 on June 30. One changelog line, five features: "add support for Managed Agents event delta streaming, agent overrides, reverse pagination, vault credential injection scoping, and agent and deployment webhook events." Two days later, v0.116.0 added the agent-memory-2026-07-22 beta header.

That's five operational API features and a memory beta in less than a week. If you've been watching the Managed Agents API, you just saw it grow up.

Source spread

• anthropic-sdk-python v0.115.0 [builder] — Primary source. All five June 30 features in one release.
• anthropic-sdk-python v0.116.0 [builder] — Agent-memory beta header, July 2.
• anthropic-sdk-python changelog [builder] — Full version history. The surrounding context shows the Managed Agents surface expanding steadily since v0.109.

What each feature actually means

There's more in that one changelog line than most API announcements ship in a quarter. Let me go through them, because they're not equally documented.

Event delta streaming is the one most immediately useful. Instead of getting a full agent event object, you get incremental deltas as the event state changes. For long-running agents this means lower latency UI — you can show progress as it happens — and potentially lower peak memory in processing pipelines that were buffering full event objects before handling them.

Agent overrides let you customize agent behavior at runtime, beyond what the initial deployment configuration allows. The specific overrideable parameters aren't published in the release notes. This one will require exploration against the API directly before you can design around it.

Reverse pagination. Small, but the right call. If you're querying agent logs or event histories, you can now paginate backward — which means "give me the most recent N events" without walking the full history forward first. Every monitoring dashboard that touches agent event logs needed this.

Vault credential injection scoping is the enterprise security story in this release. The Vault integration already let you inject credentials into agent runs without the credentials touching your application code. Scoping means you can now restrict which credentials are injectable into which agent deployments. Before this, a Vault integration was probably org-wide or nothing. Now you can lock it per deployment.

Webhook events for agents and deployments closes the operational loop. Subscribe to agent lifecycle events — started, finished, errored — and deployment events via webhooks. This is what you need to plug Managed Agents into existing observability stacks, alerting pipelines, or incident management systems. Without webhooks, you were polling or missing events.

Agent memory (beta) is the one with the most ambiguity. The SDK adds a beta request header — anthropic-beta: agent-memory-2026-07-22 — which presumably enables memory features that don't yet have stable API semantics. What exactly "memory" means in the Managed Agents context is not specified in the release notes. Could be cross-session state persistence, structured memory writes inside a session, retrieval over past sessions, or some combination. The date in the header (2026-07-22) is a beta channel identifier, not an expiration — but it implies Anthropic is planning a stable channel at some known future point.

Pros & cons

What's real:

• Vault scoping gets enterprise security teams off the blocking list. The pattern "we'll adopt Managed Agents when we can scope credential access per deployment" is common in enterprise deals. This closes that specific objection.
• Webhooks complete the operational feedback loop that was missing. Before this, you couldn't build reliable alerting on agent state changes without polling. That's table stakes for production deployments, and it's here now.
• The beta header pattern is good API design. Explicitly opting into unstable semantics beats a hidden flag or an "advanced mode" setting that changes behavior without making the instability legible.
• Reverse pagination is a small quality-of-life fix that every team was working around manually. A fix that's right.

What deserves a side-eye:

• Five features in one release, one changelog line, documentation to catch up. These are API surface changes builders will build on. The shape of some of them — agent overrides especially — matters a lot before you design an architecture that depends on them. Read the source before committing.
• "Agent overrides" is the most underspecified item here. What can you override? What's the precedence model when override conflicts with deployment config? That's important before you build a product around the feature.
• Agent memory beta is genuinely uncertain. "Agent memory" can mean different things, and what it means shapes whether you should use it for durable state. Don't build production-critical storage on a beta header.

Samwise's take

What builders need to know

Further reading

• anthropic-sdk-python v0.115.0 release — primary source for all five June 30 features
• anthropic-sdk-python v0.116.0 release — agent-memory beta header
• anthropic-sdk-python changelog — full version history showing the Managed Agents surface growing since v0.109