AI SDK Release Radar

Know what changed before you upgrade your AI stack. Track OpenAI, LangChain and LlamaIndex Python SDKs and turn official release notes into an English upgrade review with structured JSON, source links and clear evidence boundaries.

Check a single dependency or a team’s dependency list. Schedule recurring runs and connect the output to GitHub Actions, n8n, Make, Zapier or your own automation through Apify’s API and native webhooks.

What you get

• Latest stable versions from public PyPI metadata.
• Release notes between your installed version and the latest stable version.
• Breaking-change, deprecated-API, removed-feature and migration signals, with quoted official evidence.
• A risk level, update highlights and practical next steps for each dependency.
• A coverage check against PyPI’s published stable versions. Missing release notes produce unknown overall risk, rather than a reassuring guess.
• One JSON dataset row per unique dependency, a full JSON report and a readable English Markdown report.

No API key or LLM subscription is required. The MVP uses transparent, deterministic evidence rules, not a generative AI model. The name refers to the AI SDKs it monitors. Official release statements and rule-based inferences are labeled separately. It does not read or modify your application code.

Quick start

1. Keep the example input for a first run, or replace currentVersion with the exact version installed in your project.
2. Click Start.
3. Open the Dependency risk table, Full JSON report or English upgrade report in the run’s Output tab.

{
  "dependencies": [
    {"package": "openai", "currentVersion": "3.34.0"},
    {"package": "langchain", "currentVersion": "1.4.4"},
    {"package": "llama-index-core", "currentVersion": "0.14.24"}
  ]
}

The example versions are real published versions checked on October 19, 2026. Latest versions and results change over time.

Supported packages

Package	Repository selection	Notes
openai	Automatic	Official OpenAI Python SDK
langchain	Automatic	Only matching langchain== releases
langchain-core	Automatic	Only matching langchain-core== releases
langchain-openai	Automatic	Matching package tags; the package is maintained in the LangChain repository
llama-index-core	Automatic	Only the core package’s section in shared release notes
llama-index	Automatic	Requires an explicit llama-index section; otherwise the report is partial
Other public Python packages	Supply repository	Must link the repository from PyPI metadata and use v1.2.3, 1.2.3 or package==1.2.3 GitHub release tags

{
  "dependencies": [
    {
      "package": "openai",
      "repository": "openai/openai-python",
      "currentVersion": "3.34.0"
    }
  ]
}

repository also accepts https://github.com/owner/repository. Use public PyPI package names and exact, published stable versions. Version ranges, private repositories, prerelease/dev versions and local-version suffixes are not supported. Maximum: 20 input dependencies per run. Exact duplicate entries are analyzed once.

Understand the report

Field	Meaning
status	success, partial, error or skipped
riskLevel	none for no update; low, medium or high for complete release coverage; unknown when evidence is missing or a source fails
observedRiskLevel	Risk suggested by available notes; does not override unknown when coverage is incomplete
findings[].basis	official_statement for a quoted release statement; rule_inference for a major-version heuristic
findings[].evidence	Source URL, exact source excerpt and section, where available
coverage	Number of expected and reviewed versions, completeness and missing versions
billable	Whether the persisted report qualifies for an analysis event; actual charges are recorded by Apify, separately
retrievedAt	UTC timestamp of this source check

The official statement label applies to the quoted text. Categorization, potential impact and upgrade advice are rules-based interpretations. A major-version change is a heuristic, not proof that an API broke. low means no explicit risk signals were found in the reviewed notes; it does not mean a verified safe upgrade.

Example output

This shortened excerpt is from a live public-source check on October 19, 2026:

{
  "package": "llama-index-core",
  "currentVersion": "0.14.24",
  "latestVersion": "0.14.25",
  "status": "success",
  "riskLevel": "high",
  "analysisEngine": "evidence-rules-v1",
  "llmUsed": false,
  "coverage": {
    "complete": true,
    "expectedVersions": 1,
    "analyzedVersions": 1,
    "missingVersions": []
  },
  "billable": true
}

The full output also contains findings, source excerpts, recommendations, release notes, source URLs and retrieval timestamps. Export the dataset as JSON, CSV or Excel through Apify; JSON preserves nested evidence best.

Pricing

The launch price is $0.005 per complete dependency analysis ($5 per 1,000 analyses), including checks that confirm no update is available. The live Pricing tab is authoritative. There is no monthly Actor rental or LLM surcharge.

• A complete analysis requires verified public version metadata and, when an update exists, release notes for every non-yanked stable target version in the checked window.
• Errors, incomplete coverage and budget-skipped dependencies do not trigger dependency-analysis events.
• There is no startup charge or automatic per-dataset-row charge. Results are persisted before the custom analysis events are charged.
• Set Max cost per run to control spending. An input of three complete dependencies costs $0.015 at the launch price. A report with missing historical notes is returned without an analysis-event charge.
• Platform execution costs for your own development/test runs and builds are governed by your Apify account plan. Standard Store runs include platform usage in the event price. Standby mode is not supported.

Schedule and integrate

Save the input as an Apify Task, then create an Apify Schedule for a daily or weekly check. Keep currentVersion aligned with your project’s installed pin; the Actor does not change it automatically.

Resurrecting a run with an existing report repairs any missing output and does not issue another analysis event. Start a fresh run to request a new source check.

Use a native Apify webhook for ACTOR.RUN.SUCCEEDED to notify your automation, then fetch the run’s default dataset or REPORT record. A successful run may contain per-dependency error rows, so inspect status and coverage.complete before routing an upgrade alert.

import os
from apify_client import ApifyClient

client = ApifyClient(os.environ["APIFY_TOKEN"])
run = client.actor("Serendy/ai-sdk-release-radar").call(run_input={
    "dependencies": [{"package": "openai", "currentVersion": "3.34.0"}]
})
for result in client.dataset(run["defaultDatasetId"]).iterate_items():
    print(result["package"], result["status"], result["riskLevel"])

Never put API tokens in a public issue, report or source file. Apify supplies its run credentials automatically in the cloud.

Limits and troubleshooting

• Incomplete coverage: This MVP reads up to 300 recent GitHub releases per repository. Older upgrades, packages without version-specific notes and missing package sections return partial reports without analysis charges. Check a smaller upgrade window or read the missing versions manually.
• Rate limit: Public GitHub API limits apply per egress IP. The Actor returns rate_limited and does not charge that dependency. Retry later or schedule less frequently.
• Source unavailable: Transient network and server failures have bounded retries. Retry the run later. Other dependencies can still complete.
• Unknown current version / ahead of latest: Check pip show PACKAGE and enter its exact published stable version.
• Repository mismatch: Use the official repository linked from the package’s PyPI project metadata. This prevents release notes from unrelated projects entering your report.
• Budget reached: Increase the run’s maximum charge or submit fewer dependencies. Skipped dependencies are not charged.

This tool cannot guarantee compatibility or discover every undocumented breaking change. It excludes prerelease and yanked target releases, does not resolve your transitive dependency tree and does not execute packages. Review source evidence and run your application tests before upgrading.

Data and support

The Actor reads public metadata and release notes from PyPI and GitHub. Your dependency list and generated reports are stored in your run’s Apify storage under your account’s retention settings. There is no external analytics service or LLM provider in the Actor.

For help, open an issue in the Actor’s Apify Store Issues tab with the package, version, error code and a redacted input. Do not include credentials or private code.