Upgrade to SDK and CLI 2.0

The Python SDK, the Node.js SDK and the CLI each released 2.0.0 on 2026-08-26. The major version is for removals only: every call that 2.0.0 removed sent its request to an endpoint the API does not serve, so it had never returned a result. Nothing that worked was taken away, and nothing was renamed. If your code never used the calls below, upgrade with no changes. Version 1.x of each SDK gets security fixes until 2027-08-26, 12 months after 2.0.0 shipped, as Versioning & support promises.

Find what you need to change

Run this from the root of your project. No output means nothing to change.

grep -rnE '\.(completions\.create|embeddings\.create|scrape\.(url|batch)|redteam\.(run_autonomous|intelligence_stats))\(' --include='*.py' . | grep -v 'chat\.completions'

grep -rnE '\.(completions\.create|embeddings\.create|scrape\.(url|batch)|redteam\.(runAutonomous|intelligenceStats))\(|AutonomousRedTeam(Report|Request)|\bCompletion(Request|Response)|\bEmbedding(Request|Response)|IntelligenceStats|ScrapeResult' --include='*.ts' --include='*.js' --exclude-dir=node_modules . | grep -v 'chat\.completions'

grep -rnE 'promptguard (logs|events)\b|promptguard redteam .*--(autonomous|budget)' .

Python SDK

Removed from both PromptGuard and PromptGuardAsync:

Removed in 2.0.0	What to use instead
`client.completions.create(...)`	`client.chat.completions.create(...)`, which is unchanged
`client.embeddings.create(...)`	Nothing. Delete the call, or call your model provider’s embeddings API directly
`client.scrape.url(...)`, `client.scrape.batch(...)`	Nothing. Delete the call
`client.redteam.run_autonomous(...)`	Nothing equivalent. `client.redteam.run_all(...)` runs the fixed attack corpus against a preset
`client.redteam.intelligence_stats()`	Nothing. Delete the call

Text completions. Send the prompt as one user message:

Before (1.x)

from promptguard import PromptGuard

pg = PromptGuard(api_key="pg_live_xxx")
response = pg.completions.create(model="gpt-5-nano", prompt="Summarise this ticket")

After (2.0)

from promptguard import PromptGuard

pg = PromptGuard(api_key="pg_live_xxx")
response = pg.chat.completions.create(
    model="gpt-5-nano",
    messages=[{"role": "user", "content": "Summarise this ticket"}],
)

Embeddings, scraping and intelligence stats. Delete the calls. Each one raised an error on every call in 1.x, so no working code depends on what they returned:

Before (1.x)

vectors = pg.embeddings.create(model="text-embedding-3-small", input="hello")
page = pg.scrape.url("https://example.com")
stats = pg.redteam.intelligence_stats()

After (2.0)

# Removed: these calls never returned a result.

Autonomous red team. It is not offered to API keys. To test your policy against PromptGuard’s attack corpus, run the whole corpus against a preset:

Before (1.x)

report = pg.redteam.run_autonomous(budget=200, target_preset="support_bot:strict")

After (2.0)

summary = pg.redteam.run_all(target_preset="support_bot:strict")
print(f"Blocked {summary['blocked']}/{summary['total_tests']}")

Node.js SDK

Removed in 2.0.0	What to use instead
`pg.completions.create(...)`	`pg.chat.completions.create(...)`, which is unchanged
`pg.embeddings.create(...)`	Nothing. Delete the call, or call your model provider’s embeddings API directly
`pg.scrape.url(...)`, `pg.scrape.batch(...)`	Nothing. Delete the call
`pg.redteam.runAutonomous(...)`	Nothing equivalent. `pg.redteam.runAll(...)` runs the fixed attack corpus against a preset
`pg.redteam.intelligenceStats()`	Nothing. Delete the call
The exported types `AutonomousRedTeamReport`, `AutonomousRedTeamRequest`, `CompletionRequest`, `CompletionResponse`, `EmbeddingRequest`, `EmbeddingResponse`, `IntelligenceStats` and `ScrapeResult`	Delete the import. For chat, `ChatCompletionRequest` and `ChatCompletionResponse` are unchanged

The removed types are why TypeScript code that only imported them fails to compile on 2.0.0, even though it never called anything. Text completions. Send the prompt as one user message:

Before (1.x)

import { PromptGuard, type CompletionResponse } from 'promptguard-sdk';

const pg = new PromptGuard({ apiKey: process.env.PROMPTGUARD_API_KEY! });
const response: CompletionResponse = await pg.completions.create({
  model: 'gpt-5-nano',
  prompt: 'Summarise this ticket',
});

After (2.0)

import { PromptGuard, type ChatCompletionResponse } from 'promptguard-sdk';

const pg = new PromptGuard({ apiKey: process.env.PROMPTGUARD_API_KEY! });
const response: ChatCompletionResponse = await pg.chat.completions.create({
  model: 'gpt-5-nano',
  messages: [{ role: 'user', content: 'Summarise this ticket' }],
});

Autonomous red team. Run the whole corpus against a preset instead:

Before (1.x)

const report = await pg.redteam.runAutonomous({ budget: 200, targetPreset: 'support_bot:strict' });

After (2.0)

const summary = await pg.redteam.runAll('support_bot:strict');
console.log(`Blocked ${summary.blocked}/${summary.totalTests}`);

pg.embeddings, pg.scrape and pg.redteam.intelligenceStats(): delete the calls and their type imports.

CLI

Removed in 2.0.0	What to use instead
`promptguard logs`	The Activity page in the dashboard. `promptguard dashboard` opens it in your browser
`promptguard events`	The same
`promptguard redteam --autonomous` and `--budget`	`promptguard redteam`, which runs the attack corpus against a preset
The intelligence-stats output	Nothing

A script that still calls a removed command now exits with an unrecognised-subcommand error instead of a request failure, so a CI step that ignored the old failure will start failing.

Before (1.x)

promptguard logs --limit 50
promptguard redteam --autonomous --budget 50 --preset strict

After (2.0)

promptguard dashboard
promptguard redteam --preset strict

What also changed, with nothing to do

The red team calls that 2.0.0 kept (run_test, run_all and run_custom in Python, runTest, runAll and runCustom in Node.js, and promptguard redteam) got an authorization error for every API key in 1.x, because they called an admin-only path. From 2.0.0 they work with an API key that has the proxy scope. Their names and arguments did not change.

Shadow AI desktop agent 0.10

The desktop agent is versioned on its own and has no 2.0. Its one breaking change so far is in 0.10.0: protection runs only while the PromptGuard Shadow app is running. If an RMM tool or login script starts the agent with pgshadow start and no app, set the managed key headless: true before you upgrade, or protection on those machines stops. What to do for each later agent Release is on Shadow AI desktop agent upgrade notes.

Next steps

Versioning & support

How long each SDK and CLI version gets fixes

Changelog

What changed in each Release, filterable by product

Python SDK

Use the 2.x Python SDK

Node.js SDK

Use the 2.x Node.js SDK

Get Started

Paths

Core Concepts

Gateway

Integrations

Shadow AI

SDKs

Developer tools

Cookbooks

Platform

Going to Production

Resources

Roadmap / Preview

Upgrade to SDK and CLI 2.0

Find what you need to change

Python SDK

Node.js SDK

CLI

What also changed, with nothing to do

Shadow AI desktop agent 0.10

Next steps

Versioning & support

Changelog

Python SDK

Node.js SDK

​Find what you need to change

​Python SDK

​Node.js SDK

​CLI

​What also changed, with nothing to do

​Shadow AI desktop agent 0.10

​Next steps

Versioning & support

Changelog

Python SDK

Node.js SDK

Find what you need to change

Python SDK

Node.js SDK

CLI

What also changed, with nothing to do

Shadow AI desktop agent 0.10

Next steps