Fetching from the wire…
OSS2026-09-26 · source-backed
v1.0.0, published September 25 at 8,323 stars, adds percent formatting, format(), CWD support, eager awaits, long-int handling across pow/round/math, and Python 3.14t wheels. The docs say filesystem, environment and network don't exist inside the interpreter unless you pass in functions or mounts, and put startup under 1ms from a running pool against about 1,500ms for a sandbox service. For code-mode agents that write Python instead of chaining tool calls, that startup difference changes what's feasible per turn.
Each link below shares sources, entities, or timing with this story.
Pydantic's validate_call stores the caller's f_locals on the wrapper it builds, and Agno's Function caches those wrappers for the process lifetime, so on 3.13 each cached tool kept the Agent, session and run state alive indefinitely. Release notes here. The same release adds P...
PR #58613 adds thinking to /v1/messages, which had been silently dropped through Pydantic's extra='ignore'. The PR explains the concrete failure: Claude Code's client-side auto-mode classifier sends thinking disabled with max_tokens 64 and expects <severity>N immediately, but...
PR #8886, in 1.2.12, accepts a JSON Schema dict, Pydantic model, TypedDict or dataclass. Pydantic-style types validate on resume and an invalid value raises from the node before it commits. Studio renders a typed form instead of a free-form JSON box. That closes the gap where...
The v2 streaming format brings full type safety to all stream/invoke methods — strongly-typed StreamPart dicts with Pydantic model coercion. v1 remains default so nothing breaks. The langgraph-cli 0.4.14 patch simultaneously resolves three CVEs. LangGraph GitHub
LangGraph 1.1 ships opt-in version="v2" streaming that yields strongly-typed StreamPart dicts with type, namespace, data, and interrupts — eliminating the untyped dict problem that plagued production observability. Pydantic model coercion is automatic. Time-travel with interru...
Mastra core 1.70.0 added ModelSelectionProcessor, which uses a built-in classifier to pick the best or cheapest model per run, with a first-step scope and a minProbability control. Pydantic AI 2.50.0 reworked its DecisionModel so routes carry one label each and added an opt-in...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.