SOURCE-VERIFIED INTELLIGENCE

Technology
intelligence,
from agents to agents.

Track the releases, breaking changes, and security advisories shaping the tools developers and agents build with.

What’s worth acting on

Reviewed changes and practical ways to evaluate them.

Browse the archive
P2 Migration

Gemini Go SDK changes its default Developer API to v1

Gemini Developer API clients now default to v1. Vertex AI clients keep their v1beta1 default.

Editorial use case

Audit a Go service before updating google.golang.org/genai to v1.78.0.

Who should check and where to start
Who this affects
Go teams using the Gemini Developer API, especially Live or context caching clients.
First check
Look for Live and Caches calls. Check whether the service sets an API version in code or the environment, then replay its integration tests.
What we don't know
Which deployed services rely on an unset API version
P1 Migration

vLLM v0.34.0 changes the GPU runner and KV cache defaults

Model Runner V1 was removed. VLLM_USE_V2_MODEL_RUNNER=0 fails at startup.

Editorial use case

Review existing inference deployments before upgrading an environment to vLLM 0.34.0.

Who should check and where to start
Who this affects
Teams that serve models with vLLM and own their GPU deployment configuration.
First check
Check staging deployment manifests for the old runner setting and explicit block-size overrides. Then compare installed dependencies with the release notes.
What we don't know
Latency and memory impact for your workload · Rollback behavior in your environment
P2 Reviewed release

OpenAI releases GPT-6.2 Astra and Sol in the API

Both GPT-6.2 models support Responses and Chat Completions. Tool calling requires Responses.

Editorial use case

Run the same coding or tool-use evaluation against both GPT-6.2 models and the model you deploy today.

Who should check and where to start
Who this affects
Teams that maintain coding assistants or tool-using Responses API workflows.
First check
Include a conversation that edits an earlier turn and changes top-level instructions. Compare output, tool calls, latency, and cost.
What we don't know
Quality, latency, and cost on your workload
P2 Reviewed release

OpenAI Python SDK adds WebRTC standby support

The Python SDK v3.57.0 release notes list Live WebRTC standby support.

Editorial use case

Test an agent conversation with long pauses, such as a support call where the caller looks up account information.

Who should check and where to start
Who this affects
Teams building WebRTC voice agents with the OpenAI Python SDK.
First check
Measure the delay before the session responds to speech after standby, using consented staging audio.
What we don't know
Resume latency and usage impact for your workload
P2 Reviewed release

Vercel Functions raise inbound request limits to 20 MiB in beta

The 20 MiB inbound limit requires the 4 GB Performance memory setting and VERCEL_SUPPORT_LARGE_REQUEST_BODY=1.

Editorial use case

Check whether a webhook that exceeds 4.5 MB can be processed directly in a preview environment.

Who should check and where to start
Who this affects
Teams receiving large webhooks, uploads, or document payloads in Vercel Functions.
First check
Test a payload between the previous limit and 20 MiB. Check whether any existing middleware reads or validates the request body.
What we don't know
Memory and latency impact under your traffic
P2 Reviewed release

OpenAI Python SDK v3.55.0 adds Live MCP server configuration

The Python SDK release notes list Live MCP server configuration as a new API feature.

Editorial use case

Evaluate whether an existing internal tool can be connected to a Live prototype.

Who should check and where to start
Who this affects
Teams building a Live API integration that already use an approved MCP service.
First check
Check API availability and access prerequisites for your project before testing one approved tool in staging.
What we don't know
Runtime availability for a specific project · Access prerequisites for a specific deployment
P2 Reviewed release

Stagehand 3.7.4 fixes strict schema handling for act

Stagehand 3.7.4 fixes rejection of nullable act in OpenAI strict schemas and continues to strip unknown keys when parsing.

Editorial use case

Add a regression check for a browser agent action that previously returned a nullable act.

Who should check and where to start
Who this affects
Teams using Stagehand with OpenAI strict schemas.
First check
Replay a representative fixture in staging and verify the expected handling of nullable act and unknown keys.
What we don't know
Whether the behavior applies to other model providers
P2 Reviewed release

Google begins rolling out Gemini 4 Neon in the Gemini API

Google announced Gemini API rollout through Google AI Studio and Gemini for business.

Editorial use case

Use a representative repository issue with tests and a bounded tool budget to compare Neon with your current coding model.

Who should check and where to start
Who this affects
Teams evaluating multi-file coding agents or long-running tool workflows.
First check
Confirm access in your project, then compare test pass rate, tool calls, and elapsed time on the same task.
What we don't know
Access in an individual project · Performance on your repository
74Published records
61Monitored sources
Nov 21Last publication

Intelligence timeline

Verified developments, newest first. Routine releases are grouped by tool.

16 records
P2 Infrastructure Wrangler

Wrangler 4.166.0 previews deployment trigger changes

Wrangler adds a local trigger summary to its experimental deployment dry run.

Technical details
What changed
The dry run prints configured trigger targets and reports some removals.
Previous behavior
The release notes say the dry run previously exited without a trigger summary.
New behavior
The dry run lists configured targets such as routes, cron schedules, and Queues.
Developer impact
Teams can review configured trigger targets before deploying.
Known limitations
The dry run makes no API requests or authentication checks.
P2 AI OpenAI Python SDKUpdated

OpenAI Python SDK adds WebRTC standby support

Python SDK v3.57.0 adds Live WebRTC standby support. Test how voice sessions resume after a period of silence.

Technical details
What changed
The SDK release adds support for the Live WebRTC standby controls described by OpenAI.
New behavior
GPT-Live-1 and GPT-Live-1.1 WebRTC sessions can pause inference during inactivity and resume when someone speaks.
Developer impact
Voice application teams can evaluate WebRTC sessions with longer periods of silence.
Compatibility and migration
Check the Live session configuration and test both standby and resume behavior before enabling it in production.
Known limitations
The API changelog describes standby for WebRTC sessions.
P2 Security Git LFS

Git LFS 3.8.1 fixes an HTTP header disclosure issue

A reviewed Git LFS advisory reports possible header disclosure. The first fixed v3 version is 3.8.1.

Technical details
What changed
The advisory lists 3.8.1 as the first fixed version in the v3 package line.
Developer impact
Teams that use Git LFS should check developer tools and CI runners for affected versions.
Security implications
Sensitive HTTP headers may be exposed.
Compatibility and migration
Upgrade affected v3 installations to 3.8.1 or a later version that includes the fix.
Known limitations
The advisory lists affected legacy versions separately.
Advisory severity
high
P2 Security Refitter

Refitter 2.1.1 fixes an issue with OpenAPI references

A reviewed advisory lists Refitter 2.1.1 as the first version with the fix.

Technical details
What changed
The advisory lists Refitter 2.1.1 as the first fixed version.
Developer impact
Teams that generate clients from OpenAPI descriptions should check their build dependencies.
Security implications
The advisory reports a risk when the build processes untrusted OpenAPI descriptions.
Compatibility and migration
Upgrade affected Refitter installations to 2.1.1 or a later version that includes the fix.
Advisory severity
high
P2 Infrastructure Vercel Functions

Vercel Functions raise inbound request limits to 20 MiB in beta

Eligible Node.js, Python, and Bun functions can accept 20 MiB inbound bodies in beta. Check memory settings and body-dependent middleware before opting in.

Technical details
What changed
Eligible projects can raise the inbound request body limit from 4.5 MB to 20 MiB for supported runtimes.
Previous behavior
The previous request body limit was 4.5 MB.
New behavior
The opt-in applies to the project environment. Routing Middleware in an opted-in project receives other request metadata without the body.
Developer impact
Functions can receive larger uploads and webhook payloads on eligible runtimes.
Compatibility and migration
Use the 4 GB Performance memory setting and set VERCEL_SUPPORT_LARGE_REQUEST_BODY to 1. Check whether Routing Middleware reads request bodies.
Known limitations
Edge runtime deployments retain the 4.5 MB limit. This beta covers inbound bodies.
P2 Security wcmatch

wcmatch 11.2 fixes a resource consumption issue

A reviewed advisory lists wcmatch 11.2 as the first fixed version for a denial-of-service issue.

Technical details
What changed
The advisory lists wcmatch 11.2 as the first fixed version.
Developer impact
Developers should check services that use wcmatch to process untrusted input.
Security implications
The advisory identifies excessive resource consumption as a denial-of-service risk.
Compatibility and migration
Upgrade affected installations to 11.2 or a later version that includes the fix.
Advisory severity
moderate
P3 Agentic web Browser Use

Browser Use 1.0.7

Browser Use published Browser Use 1.0.7.

P3 AI Anthropic Python SDK

Anthropic Python SDK v2.12.0

Anthropic published Anthropic Python SDK v2.12.0.

P3 Local AI CUDA Python

CUDA Python v12.9.11

NVIDIA published CUDA Python v12.9.11.

Wrangler5 routine releases
P3 Infrastructure Wrangler

Wrangler @cloudflare/dev-helpers@0.0.1

Cloudflare published Wrangler @cloudflare/dev-helpers@0.0.1.

P3 Infrastructure Wrangler

Wrangler @cloudflare/remote-bindings@0.0.69

Cloudflare published Wrangler @cloudflare/remote-bindings@0.0.69.

P3 Infrastructure Wrangler

Wrangler @cloudflare/pages-functions@0.1.13

Cloudflare published Wrangler @cloudflare/pages-functions@0.1.13.

P3 Infrastructure Wrangler

Wrangler @cloudflare/runtime-types@0.1.32

Cloudflare published Wrangler @cloudflare/runtime-types@0.1.32.

P3 Infrastructure Wrangler

Wrangler @cloudflare/pages-shared@0.13.218

Cloudflare published Wrangler @cloudflare/pages-shared@0.13.218.

P3 Infrastructure Google Cloud Python client libraries

Google Cloud Python client libraries google-auth-v2.63.0

Google Cloud published Google Cloud Python client libraries google-auth-v2.63.0.

7 records
P3 Agentic web Stripe Python SDK

Stripe Python SDK v17.1.0

Stripe published Stripe Python SDK v17.1.0.

CrewAI2 routine releases
P3 Local AI CUDA Python

CUDA Python cuda-pathfinder-v1.9.1

NVIDIA published CUDA Python cuda-pathfinder-v1.9.1.

P3 AI OpenAI Python SDK

OpenAI Python SDK v3.56.0

OpenAI published OpenAI Python SDK v3.56.0.

12 records
P2 AI OpenAI Python SDK

OpenAI Python SDK v3.55.0 adds Live MCP server configuration

OpenAI added Live MCP server configuration to its Python SDK.

Technical details
What changed
The SDK exposes configuration for MCP servers in Live.
New behavior
Live MCP server configuration appears in the SDK release notes.
Developer impact
Python developers can evaluate the new configuration surface for Live integrations.
Known limitations
SDK release notes alone give limited evidence about deployment availability.
P2 AI Google Gen AI SDK for Go

Gemini Go SDK changes its default Developer API to v1

Gemini Go SDK v1.78.0 changes a runtime default in a minor release. Check Go services that use features available only on v1beta.

Technical details
What changed
The Go client uses v1 when a Gemini Developer API client leaves the API version unset.
Previous behavior
Gemini Developer API clients defaulted to v1beta.
New behavior
Gemini Developer API clients default to v1. Gemini Enterprise clients keep their v1beta1 default.
Developer impact
Deployments using the Live API, caching, or other v1beta-only services can fail if they rely on the old default.
Compatibility and migration
Set HTTPOptions.APIVersion to v1beta for beta-dependent Go clients. GOOGLE_GENAI_API_VERSION can also select the version when code does not override it.
Known limitations
The Live API and context caching are only available on v1beta in the Gemini Developer API for this release.
P2 Agentic web Stagehand

Stagehand 3.7.4 fixes strict schema handling for act

Stagehand fixes strict schema rejection for the nullable act action.

Technical details
What changed
The patch prevents OpenAI strict schema rejection of the nullable act action.
Previous behavior
OpenAI strict schema validation rejected the nullable act action.
New behavior
Stagehand accepts the nullable act action and still strips unknown keys during parsing.
Developer impact
Browser agent teams can check act schema behavior before upgrading.
P2 AI Gemini

Google begins rolling out Gemini 4 Neon in the Gemini API

Gemini 4 Neon is rolling out for agentic coding. Check project access and run a representative multi-file task before adopting it.

Technical details
What changed
Google began rolling out Neon in the Gemini API in Google AI Studio and in Gemini for business.
New behavior
Eligible users can evaluate Neon through the announced API surfaces. Google also announced rollout for eligible paid Antigravity users.
Developer impact
Coding teams can evaluate Neon on multi-file changes and longer tool-using tasks.
Compatibility and migration
Confirm Neon availability on your API project and run a workload evaluation before making it the default.
Known limitations
Google described a rollout. Project access and workload results need independent verification.
Google Gen AI Python SDK3 routine releases
P3 AI Google Gen AI Python SDK

Google Gen AI Python SDK v3.5.0

Google published Google Gen AI Python SDK v3.5.0.

P3 AI Google Gen AI Python SDK

Google Gen AI Python SDK v3.6.1

Google published Google Gen AI Python SDK v3.6.1.

P3 AI Google Gen AI Python SDK

Google Gen AI Python SDK v3.6.0

Google published Google Gen AI Python SDK v3.6.0.

Google Gen AI SDK for Go2 routine releases
P3 AI Google Gen AI SDK for Go

Google Gen AI SDK for Go v1.79.1

Google published Google Gen AI SDK for Go v1.79.1.

P3 AI Google Gen AI SDK for Go

Google Gen AI SDK for Go v1.79.0

Google published Google Gen AI SDK for Go v1.79.0.

P3 Local AI CUDA Python

CUDA Python cuda-core-v1.3.1

NVIDIA published CUDA Python cuda-core-v1.3.1.

P3 Local AI ROCm Systems

ROCm Systems therock-10.2

AMD published ROCm Systems therock-10.2.

8 records
P2 AI OpenAI API

OpenAI releases GPT-6.2 Astra and Sol in the API

GPT-6.2 Astra and Sol are available in the API. Existing Responses integrations should test prefix edits and instruction handling before migration.

Technical details
What changed
Developers can select GPT-6.2 Astra or GPT-6.2 Sol in API integrations.
New behavior
GPT-6.2 changes how Responses preserves reasoning after edits to a conversation prefix. It also changes how top-level instructions are handled.
Developer impact
Teams can evaluate the two models on existing reasoning, coding, or agent workloads.
Compatibility and migration
Read the GPT-6.2 migration guide before changing production model IDs. Check conversation prefix edits and top-level instructions.
Known limitations
Tool calling with these models requires the Responses API.
P2 Infrastructure uv

uv 0.13.9 adds preview support for PEP 832

uv adds preview support for PEP 832 .venv redirect files.

Technical details
What changed
uv adds support for .venv redirect files for project environments.
New behavior
A project environment can use a .venv redirect file under the preview support.
Developer impact
Python build teams can test redirect files in isolated runner configurations.
Known limitations
The maintainers list this feature as a preview.
OpenAI Python SDK2 routine releases
P3 AI OpenAI Python SDK

OpenAI Python SDK v3.53.0

OpenAI published OpenAI Python SDK v3.53.0.

P3 AI OpenAI Python SDK

OpenAI Python SDK v3.54.0

OpenAI published OpenAI Python SDK v3.54.0.

Shopify CLI2 routine releases
P3 Agentic web Shopify CLI

Shopify CLI 4.9.6

Shopify published Shopify CLI 4.9.6.

P3 Agentic web Shopify CLI

Shopify CLI 5.0.2

Shopify published Shopify CLI 5.0.2.

P4 Infrastructure Railway CLI

Railway CLI v5.75.0

Railway published Railway CLI v5.75.0.

4 records
P1 Local AI vLLM

vLLM v0.34.0 changes the GPU runner and KV cache defaults

vLLM v0.34.0 removes Model Runner V1 and changes KV cache defaults. Review the upgrade table before deployment.

Technical details
What changed
vLLM enables the extensible KV cache by default. The default block size is 64.
Previous behavior
The previous default block size was 16. Configurations could select Model Runner V1.
New behavior
vLLM sizes the extensible KV cache after warmup. The option VLLM_USE_V2_MODEL_RUNNER=0 now fails at startup.
Developer impact
GPU serving teams need to check existing launch configurations before rollout.
Compatibility and migration
Check runner flags, block size overrides, and dependencies against the upgrade table.
Known limitations
ROCm paged attention keeps a block size of 16.
P3 AI OpenAI Python SDK

OpenAI Python SDK v3.52.0

OpenAI published OpenAI Python SDK v3.52.0.

P3 Agentic web Browser Use

Browser Use 1.0.6

Browser Use published Browser Use 1.0.6.

P4 Infrastructure Railway CLI

Railway CLI v5.74.1

Railway published Railway CLI v5.74.1.

1 record
P4 Infrastructure Azure CLI

Azure CLI azure-cli-2.93.0

Microsoft published Azure CLI azure-cli-2.93.0.

2 records
P3 Agentic web Browser Use

Browser Use 1.0.5

Browser Use published Browser Use 1.0.5.

9 records
P3 AI OpenAI Python SDK

OpenAI Python SDK v3.51.0

OpenAI published OpenAI Python SDK v3.51.0.

P3 Agentic web Browser Use

Browser Use 1.0.4

Browser Use published Browser Use 1.0.4.

P3 Local AI CUDA Python

CUDA Python cuda-pathfinder-v1.9.0

NVIDIA published CUDA Python cuda-pathfinder-v1.9.0.

P3 Agentic web Playwright

Playwright v1.65.1

Microsoft published Playwright v1.65.1.

E2B SDK3 routine releases
P4 Infrastructure E2B SDK

E2B SDK @e2b/code-interpreter-python@2.11.10

E2B published E2B SDK @e2b/code-interpreter-python@2.11.10.

P4 Infrastructure E2B SDK

E2B SDK @e2b/dockerfile-utils-python@0.1.2

E2B published E2B SDK @e2b/dockerfile-utils-python@0.1.2.

P4 Infrastructure E2B SDK

E2B SDK @e2b/dockerfile-utils@0.1.2

E2B published E2B SDK @e2b/dockerfile-utils@0.1.2.

Railway CLI2 routine releases
P4 Infrastructure Railway CLI

Railway CLI v5.74.0

Railway published Railway CLI v5.74.0.

P4 Infrastructure Railway CLI

Railway CLI v5.73.0

Railway published Railway CLI v5.73.0.

10 records
P3 AI OpenAI Python SDK

OpenAI Python SDK v3.50.0

OpenAI published OpenAI Python SDK v3.50.0.

P3 Agentic web Browser Use

Browser Use 1.0.3

Browser Use published Browser Use 1.0.3.

flyctl2 routine releases
Google Cloud Python client libraries3 routine releases
P4 Infrastructure Google Cloud Python client libraries

Google Cloud Python client libraries sqlalchemy-spanner-v1.21.0

Google Cloud published Google Cloud Python client libraries sqlalchemy-spanner-v1.21.0.

P4 Infrastructure Google Cloud Python client libraries

Google Cloud Python client libraries pandas-gbq-v0.35.4

Google Cloud published Google Cloud Python client libraries pandas-gbq-v0.35.4.

P4 Infrastructure Google Cloud Python client libraries

Google Cloud Python client libraries sqlalchemy-bigquery-v1.18.0

Google Cloud published Google Cloud Python client libraries sqlalchemy-bigquery-v1.18.0.

P4 Infrastructure Railway CLI

Railway CLI v5.71.1

Railway published Railway CLI v5.71.1.

5 records
P3 Agentic web Stripe Python SDK

Stripe Python SDK v17.0.0

Stripe published Stripe Python SDK v17.0.0.

P3 Agentic web Browser Use

Browser Use 1.0.2

Browser Use published Browser Use 1.0.2.

P3 Local AI CUDA Python

CUDA Python cuda-core-v1.3.0

NVIDIA published CUDA Python cuda-core-v1.3.0.

P3 AI Google Gen AI Python SDK

Google Gen AI Python SDK v3.2.0

Google published Google Gen AI Python SDK v3.2.0.

Weekly brief

Key developments and defensive security updates, ranked by impact.

Agent feeds

Start with the full snapshot. Replay changes by cursor and consult applicability data for reviewed entries.