Veqen SDKs

https://github.com/keyveve/veqen-sdks
Admin
Scale AI inference without losing track of the bill. Free Python and JavaScript tools to see what AI...
Tokens:99,674
Snippets:563
Trust Score:3.6
License:MIT
Update:4 hours ago
Show doc for...
Tokens:
Raw
### Development setup and test commands

Source: https://github.com/keyveve/veqen-sdks/blob/main/README.md

Commands to install the Python package in editable mode with test dependencies, run the Python test suite, install JavaScript dependencies, and run JS tests. Requires Python and Node.js environments.

```sh
python -m pip install -e 'python[test]'
python -m pytest python/tests
cd js
npm ci --ignore-scripts
npm test
```

--------------------------------

### Run locally with npm commands

Source: https://github.com/keyveve/veqen-sdks/blob/main/examples/vercel-ai-cost-starter/README.md

Install dependencies, run tests, and start the development server. Requires Node.js 22 or later. After starting, open http://localhost:3000.

```sh
npm ci
npm test
npm run dev
```

--------------------------------

### Install Python SDK with pip

Source: https://github.com/keyveve/veqen-sdks/blob/main/README.md

Install the Veqen Python SDK version 1.x. Requires Python 3.10+ and has no runtime dependencies.

```sh
pip install 'veqen>=1,<2'
```

--------------------------------

### Install JavaScript SDK with npm

Source: https://github.com/keyveve/veqen-sdks/blob/main/README.md

Install the Veqen JavaScript SDK version 1.x. Requires Node.js 20+, ESM, TypeScript declarations, and no runtime dependencies.

```sh
npm install @veqen/sdk@^1
```

--------------------------------

### Register VeqenLogger with LiteLLM

Source: https://github.com/keyveve/veqen-sdks/blob/main/python/README.md

Install veqen[litellm]>=1.1.0 and register the VeqenLogger as a LiteLLM callback. Call veqen.init() optionally with VEQEN_API_KEY for the hosted ledger. The logger copies allowlisted cost metadata, not prompts or responses, and observes completed calls without enforcing pre-call budgets.

```python
import litellm
import veqen
from veqen.integrations.litellm_logger import VeqenLogger

veqen.init()  # optional VEQEN_API_KEY for the hosted ledger
litellm.callbacks = [VeqenLogger()]
```

--------------------------------

### Usage Example

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/python-integrations.md

Example showing how to register the VeqenTracingProcessor with the OpenAI Agents SDK.

```python
from agents import add_trace_processor
from veqen.integrations.openai_agents import VeqenTracingProcessor

add_trace_processor(VeqenTracingProcessor(feature="support-agent"))

```

--------------------------------

### Install veqen-pricebook and @veqen/pricebook

Source: https://github.com/keyveve/veqen-sdks/blob/main/pricebook-packages/js/README.md

Install the Python and npm packages. No runtime dependencies, no network requests, no account or API key required.

```bash
pip install veqen-pricebook
npm install @veqen/pricebook
```

--------------------------------

### Example usage of veqen.flush

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/python-core.md

Example showing how to initialize Veqen and flush events before shutdown, with a warning if some events were not uploaded.

```python
import veqen

veqen.init(api_key=os.environ["VEQEN_API_KEY"])
# ... request handling ...
ok = veqen.flush(timeout=3.0)
if not ok:
    log.warning("some Veqen events were not uploaded before shutdown")
```

--------------------------------

### Install @veqen/sdk with npm

Source: https://github.com/keyveve/veqen-sdks/blob/main/js/README.md

Install the veqen SDK package. Requires Node.js 20+, ESM, no dependencies, TypeScript types included.

```bash
npm install @veqen/sdk
```

--------------------------------

### Install veqen with pip

Source: https://github.com/keyveve/veqen-sdks/blob/main/python/README.md

Install the base veqen package. Requires Python 3.10+ and works offline with no dependencies. Optional integrations include veqen[langchain], veqen[litellm], veqen[llamaindex], and veqen[haystack] (Python 3.11+ for Haystack).

```bash
pip install veqen
```

--------------------------------

### Wrap OpenAI client and tag requests in JavaScript

Source: https://github.com/keyveve/veqen-sdks/blob/main/README.md

Wrap an OpenAI client with wrap and use withTags to attribute requests to a customer. Works locally without an account.

```javascript
import OpenAI from "openai";
import { wrap, withTags } from "@veqen/sdk";

const client = wrap(new OpenAI(), { feature: "support-bot" });
await withTags({ customer: "acme" }, () =>
  client.responses.create({ model: "gpt-5-mini", input: "Where is my order?" }));
```

--------------------------------

### Install langchain-veqen and langchain-openai

Source: https://github.com/keyveve/veqen-sdks/blob/main/integrations/langchain/README.md

Install the Veqen LangChain integration and the OpenAI chat model package. Required before using the callback handler.

```bash
pip install langchain-veqen langchain-openai
```

--------------------------------

### Usage example with withTags and wrap

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/js-core.md

Shows how to set a budget and use it with withTags and wrap.

```js
import { budget, withTags, wrap } from "@veqen/sdk";

budget(5, { customer: "*" });
await withTags({ customer: "acme" }, () => wrappedClient.chat.completions.create({ ... }));
```

--------------------------------

### handleLLMStart

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/js-wrap.md

Callback invoked when an LLM starts. Delegates to handleChatModelStart to perform the same budget check and metadata storage.

```APIDOC
## handleLLMStart

### Description
Callback invoked when an LLM starts. Delegates to handleChatModelStart to perform the same budget check and metadata storage.

### Method
`handleLLMStart(llm, prompts, runId, parentRunId, extra, tags, metadata)`

### Parameters
- `llm` - The language model instance.
- `prompts` - The prompts sent to the model.
- `runId` (string) - The run identifier.
- `parentRunId` - The parent run identifier.
- `extra` - Additional data.
- `tags` - Tags associated with the run.
- `metadata` (object) - Metadata for the run.

### Behavior
- Calls `handleChatModelStart` with the same arguments.
```

--------------------------------

### Wrap OpenAI client and tag requests in Python

Source: https://github.com/keyveve/veqen-sdks/blob/main/README.md

Wrap an OpenAI client with veqen.wrap and use veqen.tags to attribute requests to a customer. Works locally without an account.

```python
import veqen
from openai import OpenAI

client = veqen.wrap(OpenAI(), feature="support-bot")
with veqen.tags(customer="acme"):
    client.responses.create(model="gpt-5-mini", input="Where is my order?")
```

--------------------------------

### veqen init [dir]

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/js-cli.md

Initializes the current project by inspecting package.json, requirements.txt, and pyproject.toml to detect AI SDKs and prints setup instructions. Always prints the VEQEN_API_KEY note.

```APIDOC
## veqen init [dir]

### Description
Prints the setup for the current project by inspecting `package.json` (detects the Vercel AI SDK `ai`, `openai`, `@anthropic-ai/sdk`, `@google/genai`), `requirements.txt` and `pyproject.toml` (detects `anthropic`, `google-genai`, `langchain`, `litellm`, `openai-agents`, otherwise defaults to `OpenAI()`). Always prints the `VEQEN_API_KEY` note.

### Method
CLI command

### Endpoint
`veqen init [dir]`

### Parameters
#### Path Parameters
- **dir** (string) - Optional - Directory to initialize (defaults to current directory)

### Request Example
```bash
npx @veqen/sdk@1 init
```

### Response
#### Success Response
Prints setup instructions and the VEQEN_API_KEY note.

```

--------------------------------

### providerForUrl usage example

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/js-pricing.md

Example usage of providerForUrl with sample URLs.

```javascript
import { providerForUrl } from "@veqen/sdk";

providerForUrl("https://api.together.xyz/v1"); // "together"
providerForUrl("https://api.openai.com/v1");   // "openai"
```

--------------------------------

### PriceBook usage example

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/python-pricing.md

Shows how to load the price book, access its version, resolve a provider from a URL, and compute a cost for a model with usage. The example prints the version and provider, then calculates the cost for a specific model.

```python
import veqen

book = veqen.price_book()
print(book.version)                                    # 2026-10-23
print(book.provider_for_url("https://api.together.xyz"))  # together
usd = book.cost("meta-llama/Llama-3.3-70B-Instruct-Turbo",
                veqen.Usage(input_tokens=1000, output_tokens=250), provider="together")
```

--------------------------------

### Example: record with tags

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/js-core.md

Example usage of record with provider 'anthropic' and model 'claude-sonnet-5-5', passing inputTokens and outputTokens, and tags. Logs cost_usd and price_version.

```javascript
import { record } from "@veqen/sdk";

const event = record("anthropic", "claude-sonnet-5-5", {
  inputTokens: 1200, outputTokens: 300,
}, { tags: { customer: "acme" } });
console.log(event.cost_usd, event.price_version);
```

--------------------------------

### Python: veqen_pricebook.load usage example

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/pricebook-packages.md

Example of using the load function to access the price-book snapshot. Prints the version, schema, and standard rates for a model.

```python
from veqen_pricebook import load

book = load()
print(book["version"], book["schema"])                                   # 2026-10-23 2
print(book["models"]["openai/gpt-4o"]["rates"]["standard"])              # {'in': 2.5, 'out': 10.0}
```

--------------------------------

### Build and install the local package

Source: https://github.com/keyveve/veqen-sdks/blob/main/vscode-extension/README.md

Run these commands from the extension directory using Node.js 22 to install dependencies, prepare data, run tests, and package the VSIX. After packaging, install the generated VSIX via 'Extensions: Install from VSIX…' in VS Code.

```sh
npm ci --ignore-scripts
npm run prepare-data
npm test
npm run package
```

--------------------------------

### Usage example for flush with init

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/js-core.md

Shows how to initialize the SDK with an API key and flush events before a serverless function returns. If flush returns false, events remain queued for the next interval.

```javascript
import { init, flush } from "@veqen/sdk";

init({ apiKey: process.env.VEQEN_API_KEY });
// ... work ...
const ok = await flush();
if (!ok) console.warn("events remain queued for the next interval");
```

--------------------------------

### lookup usage example

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/js-pricing.md

Example usage of lookup with a model and provider.

```javascript
import { lookup } from "@veqen/sdk";

const price = lookup("gpt-4o-mini");
console.log(price.provider, price.model, price.retires ?? "no retirement");
lookup("us.anthropic.claude-sonnet-4-5-20250929-v1:0", "bedrock");
```

--------------------------------

### init with ledgerFile and environment

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/js-core.md

Example usage of init with a ledger file and environment. The ledger file appends one JSON line per event; the parent directory is created automatically.

```js
import { init } from "@veqen/sdk";

init({ ledgerFile: ".veqen/ledger.jsonl", environment: "production" });
```

--------------------------------

### Veqen SDK CLI commands for scanning, diffing, auditing, and reporting

Source: https://github.com/keyveve/veqen-sdks/blob/main/README.md

Run these commands from a terminal to scan model references for prices and retirement dates, compare model changes, import provider usage locally with your admin key, or report on a local ledger. The cost-review Action adds a pull-request summary and optional retirement checks; it never executes the reviewed application's code. Use `pull_request`, not `pull_request_target`.

```sh
npx @veqen/sdk@1 scan
npx @veqen/sdk@1 diff origin/main
npx @veqen/sdk@1 audit
npx @veqen/sdk@1 report
```

--------------------------------

### Example: record with explicit cost

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/js-core.md

Example usage of record for a self-hosted model with explicit costUsd, since the price book doesn't know it. Includes tags for feature and region.

```javascript
import { record } from "@veqen/sdk";

// Self-hosted model with an explicit cost (price book doesn't know it)
record("self-hosted", "my-vllm-70b",
  { inputTokens: 90_000, outputTokens: 4_000 },
  { costUsd: 0.0042, tags: { feature: "rag", region: "us-east" } });
```

--------------------------------

### init with apiKey and flush in serverless handler

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/js-core.md

Example showing init with an API key and custom flush interval, then flushing manually in a serverless handler. The flush returns true when the queue is empty.

```js
import { init, flush } from "@veqen/sdk";

init({ apiKey: process.env.VEQEN_API_KEY, flushIntervalMs: 1000, budgetPollMs: 15_000 });

export async function handler() {
  // ... model calls ...
  const drained = await flush(); // true when the queue is empty
  return { drained };
}
```

--------------------------------

### init(options?)

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/js-core.md

Initializes the Veqen SDK with configuration options. Sets the API key, endpoint, environment, ledger file, and polling intervals. Starts a periodic flush timer when an API key is provided.

```APIDOC
## `init(options?)`

### Description
Initializes the Veqen SDK with configuration options. Sets the API key, endpoint, environment, ledger file, and polling intervals. Starts a periodic flush timer when an API key is provided.

### Method
Function call

### Signature
```ts
export declare function init(options?: {
  apiKey?: string;
  endpoint?: string;
  environment?: string;
  ledgerFile?: string;
  flushIntervalMs?: number;
  budgetPollMs?: number;
}): void;
```

### Parameters
#### Options Object
- **apiKey** (string) - Optional - API key for authentication. Defaults to env `VEQEN_API_KEY` or `""`. When set, a `setInterval(flush, flushIntervalMs)` sender starts and `beforeExit` flushes.
- **endpoint** (string) - Optional - Base URL for the API. Defaults to env `VEQEN_ENDPOINT` or `"https://veqen-app.com"`. Trailing `/` is stripped.
- **environment** (string) - Optional - Default environment value on every event. Defaults to env `VEQEN_ENVIRONMENT`.
- **ledgerFile** (string) - Optional - Path to a file for appending one JSON line per event. Parent directory is created (failure logs a warning). Defaults to env `VEQEN_LEDGER_FILE`.
- **flushIntervalMs** (number) - Optional - Sender interval when an API key is set. Defaults to `2000`.
- **budgetPollMs** (number) - Optional - Minimum delay between `GET /v1/budgets/status` polls. Defaults to `30000`.

### Returns
`undefined`. Calling `init` again replaces the previous timer.

### Example
```js
import { init } from "@veqen/sdk";

init({ ledgerFile: ".veqen/ledger.jsonl", environment: "production" });
```

```js
import { init, flush } from "@veqen/sdk";

init({ apiKey: process.env.VEQEN_API_KEY, flushIntervalMs: 1000, budgetPollMs: 15_000 });

export async function handler() {
  // ... model calls ...
  const drained = await flush(); // true when the queue is empty
  return { drained };
}
```
```

--------------------------------

### cost usage example with valid model

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/js-pricing.md

Example of calling cost with a known model and usage, showing the returned cost. The second call includes provider and variant options.

```javascript
import { cost } from "@veqen/sdk";

cost("gpt-4o", { inputTokens: 1000, outputTokens: 500 });                    // 0.0075
cost("veo-3.1-generate-001", { units: 8 }, { provider: "google", variant: "1080p" });
```

--------------------------------

### handleChatModelStart

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/js-wrap.md

Callback invoked when a chat model starts. Stores metadata for the run and performs a budget check using the handler's tags merged with currentTags().

```APIDOC
## handleChatModelStart

### Description
Callback invoked when a chat model starts. Stores metadata for the run and performs a budget check using the handler's tags merged with currentTags().

### Method
`handleChatModelStart(_llm, _messages, runId, _parentRunId, _extra, _tags, metadata)`

### Parameters
- `_llm` - The language model instance.
- `_messages` - The messages sent to the model.
- `runId` (string) - The run identifier.
- `_parentRunId` - The parent run identifier.
- `_extra` - Additional data.
- `_tags` - Tags associated with the run.
- `metadata` (object) - Metadata for the run, stored for later use.

### Behavior
- Stores the metadata in the `runs` map keyed by `runId`.
- Calls `check({ ...this.tags, ...currentTags() })` to enforce budget limits.
```

--------------------------------

### Invalid budget usage example

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/js-core.md

Demonstrates catching the error when multiple scopes are provided.

```js
import { budget } from "@veqen/sdk";

try {
  budget(1, { customer: "acme", user: "u_1" });
} catch (error) {
  console.error(error.message); // Use one of customer/feature/user/project, ...
}
```

--------------------------------

### veqen CLI commands

Source: https://github.com/keyveve/veqen-sdks/blob/main/js/README.md

CLI commands for setup, scanning models, diffing cost impact, auditing, and reporting. Use --json, --fail-within 30, and --format markdown|json options as shown.

```bash
npx @veqen/sdk init                  # the setup for this project
```

```bash
npx @veqen/sdk scan                  # models in your code: prices, retirement dates, cheaper options (--json, --fail-within 30)
```

```bash
npx @veqen/sdk diff origin/main      # cost impact of model changes in this branch (--format markdown|json)
```

```bash
npx @veqen/sdk audit                 # last 30 days from your OpenAI or Anthropic admin key, locally
```

```bash
npx @veqen/sdk report                # cost by customer, feature and model from a local ledger (init({ ledgerFile }))
```

--------------------------------

### Usage example with veqen.cost

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/python-pricing.md

Shows constructing a Usage object and accessing prompt_tokens and tokens properties. Also calls veqen.cost with input and output tokens.

```python
import veqen

usage = veqen.Usage(input_tokens=1000, output_tokens=500, cache_read_tokens=800)
print(usage.prompt_tokens)  # 1800
print(usage.tokens)         # 2300
veqen.cost("gpt-4o", usage.input_tokens, usage.output_tokens)
```

--------------------------------

### Install and configure OpenLLMetry environment variables

Source: https://github.com/keyveve/veqen-sdks/blob/main/examples/openllmetry/README.md

Install the traceloop-sdk and OpenAI packages, then set required environment variables for OpenLLMetry. On PowerShell, use $env:NAME = 'value' syntax for these variables.

```sh
pip install traceloop-sdk==0.62.4 openai
export TRACELOOP_BASE_URL=http://127.0.0.1:4318
export TRACELOOP_TRACE_CONTENT=false
export TRACELOOP_TELEMETRY=false
export TRACELOOP_METRICS_ENABLED=false
export TRACELOOP_LOGGING_ENABLED=false
```

--------------------------------

### init function implementation (js/src/core.js:49)

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/js-core.md

Initializes the Veqen SDK state with options or environment variables. Creates the ledger directory if needed, starts a flush interval when an API key is set, and registers a beforeExit flush. Calling init again replaces the previous timer.

```js
export function init(options = {}) {
  state.apiKey = options.apiKey ?? env("VEQEN_API_KEY") ?? "";
  state.endpoint = (options.endpoint ?? env("VEQEN_ENDPOINT") ?? state.endpoint).replace(/\/$/, "");
  state.environment = options.environment ?? env("VEQEN_ENVIRONMENT");
  state.ledgerFile = options.ledgerFile ?? env("VEQEN_LEDGER_FILE");
  if (state.ledgerFile) {
    try {
      mkdirSync(dirname(state.ledgerFile), { recursive: true });
    } catch (error) {
      console.warn(`[veqen] can't create the ledger folder: ${error.message}`); // e.g. a read-only filesystem
    }
  }
  state.budgetPollMs = options.budgetPollMs ?? state.budgetPollMs;
  state.configured = true;
  clearInterval(state.timer);
  state.timer = null;
  if (state.apiKey) {
    state.timer = setInterval(() => void flush(), options.flushIntervalMs ?? 2000);
    state.timer.unref?.();
    globalThis.process?.once?.("beforeExit", () => void flush());
  }
}
```

--------------------------------

### VeqenTracingProcessor.on_trace_start

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/python-integrations.md

Called when a trace starts. This implementation is a no-op.

```APIDOC
## `VeqenTracingProcessor.on_trace_start`

### Description
Called when a trace starts. This implementation is a no-op.

### Method
Method

### Signature
```python
on_trace_start(self, trace: Any) -> None
```

### Parameters
- **trace** (Any) - Required - The trace object that started.

```

--------------------------------

### Serverless init with API key and custom polling

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/python-core.md

Initializes Veqen with an API key from environment variables, enabling event uploads and starting the background sender thread. Uses faster flush and budget polling intervals for serverless hot paths.

```python
import os
import veqen

veqen.init(
    api_key=os.environ["VEQEN_API_KEY"],
    endpoint=os.environ.get("VEQEN_ENDPOINT", "https://veqen-app.com"),
    flush_interval=1.0,   # batch faster in a serverless hot path
    budget_poll=15.0,     # poll dashboard budgets more often
)
```

--------------------------------

### Using priceBook to access version and model rates

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/js-pricing.md

Example of importing priceBook and accessing the version and a model's standard input rate.

```javascript
import { priceBook } from "@veqen/sdk";

console.log(priceBook.version);
console.log(priceBook.models["openai/gpt-4o"].rates.standard.in); // 2.5
```

--------------------------------

### Wrap an OpenAI client and tag calls with veqen

Source: https://github.com/keyveve/veqen-sdks/blob/main/python/README.md

Wrap an OpenAI client with veqen.wrap() to track costs per feature. Use veqen.tags() to attribute calls to a customer and user. Every call is priced from provider-reported usage; prompts and responses are never read.

```python
import veqen
from openai import OpenAI

client = veqen.wrap(OpenAI(), feature="support-bot")

with veqen.tags(customer="acme", user="u_42"):
    client.responses.create(model="gpt-5-mini", input="Where is my order?")
```

--------------------------------

### Python: veqen_pricebook.load — load price-book snapshot

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/pricebook-packages.md

Loads the installed, dated price-book snapshot without network access. Returns an independent copy of the dict parsed from the bundled prices.json resource. Requires Python ≥ 3.10.

```python
"""Load the installed, dated price-book snapshot without network access."""
import json
from importlib.resources import files


def load() -> dict:
    """Return an independent copy. Token rates are USD per million tokens."""
    return json.loads(files(__package__).joinpath("prices.json").read_text(encoding="utf-8"))
```

--------------------------------

### veqen_pricebook.load

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/pricebook-packages.md

Loads the installed, dated price-book snapshot without network access. Returns an independent copy of the bundled prices.json data.

```APIDOC
## veqen_pricebook.load

### Description
Loads the installed, dated price-book snapshot without network access. Returns an independent copy of the bundled prices.json data.

### Method
Function call

### Endpoint
`veqen_pricebook.load()`

### Parameters
None

### Returns
- **dict** - A fresh dictionary (independent copy) parsed from the bundled `prices.json` resource. Token rates are USD per million tokens.

### Example
```python
from veqen_pricebook import load

book = load()
print(book["version"], book["schema"])                                   # 2026-10-23 2
print(book["models"]["openai/gpt-4o"]["rates"]["standard"])              # {'in': 2.5, 'out': 10.0}
```
```

--------------------------------

### init function signature

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/python-core.md

Configures Veqen. Without an API key, costs are computed locally and written to ledger_file if set; nothing is sent. Reads VEQEN_API_KEY, VEQEN_ENDPOINT, VEQEN_ENVIRONMENT, and VEQEN_LEDGER_FILE when arguments are omitted. Setting a non-empty api_key starts a daemon thread named veqen-sender and registers atexit.register(self.flush, 2.0). Calling init again re-configures the singleton.

```python
def init(
    api_key: str | None = None,
    *,
    endpoint: str | None = None,
    environment: str | None = None,
    ledger_file: str | None = None,
    flush_interval: float = 2.0,
    budget_poll: float = 30.0,
) -> None:
    """Configure Veqen. Without an API key, costs are computed locally (and written to ``ledger_file``
    if set) and nothing is sent anywhere. Reads VEQEN_API_KEY, VEQEN_ENDPOINT, VEQEN_ENVIRONMENT and
    VEQEN_LEDGER_FILE when arguments are omitted."""
```

--------------------------------

### veqen diff origin/main --format markdown

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/js-cli.md

Compares the current branch against origin/main, reporting model changes in Markdown format. The output starts with <!-- veqen-cost-review --> for GitHub Action comment deduplication.

```bash
npx @veqen/sdk@1 diff origin/main --format markdown
```

--------------------------------

### withTags usage with wrap and OpenAI client

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/js-core.md

Example showing how to use `withTags` with a wrapped OpenAI client. The tags `{ customer: "acme", user: "u_42" }` are applied to the `responses.create` call inside the function.

```js
import { wrap, withTags } from "@veqen/sdk";

const client = wrap(new OpenAI(), { feature: "support-bot" });
await withTags({ customer: "acme", user: "u_42" }, () =>
  client.responses.create({ model: "gpt-5-mini", input: "Where is my order?" }));
```

--------------------------------

### Usage example with ChatOpenAI

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/python-integrations.md

Shows how to attach VeqenCallbackHandler to a ChatOpenAI call, both at construction and via config callbacks. The handler merges tags with current_tags() and records usage after the call.

```python
from veqen.integrations.langchain import VeqenCallbackHandler

llm = ChatOpenAI(model="gpt-4o-mini", callbacks=[VeqenCallbackHandler(feature="research-agent")])
answer = llm.invoke("Summarize this document.", config={"callbacks": [VeqenCallbackHandler(customer="acme")]})
veqen.flush()
```

--------------------------------

### Lookup and cost for video model with variant

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/python-pricing.md

Looks up a price for 'veo-3.1-generate-001' from provider 'google', gets unit rate for '1080p', and computes cost for 8 units with residency multipliers.

```python
import veqen

price = veqen.lookup("veo-3.1-generate-001", "google")
per_second = price.unit_rate("1080p")   # resolution variant
usd = price.cost(veqen.Usage(units=8), variant="1080p",
                 multipliers=veqen.price_book().residency.get("google"))
```

--------------------------------

### veqen CLI help text

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/js-cli.md

Displays the full command list for the Veqen CLI, including init, scan, diff, audit, and report subcommands with their options and descriptions.

```text
veqen — the cost ledger for AI inference

  veqen init                     How to add Veqen to this project
  veqen scan [dir] [--json] [--fail-within DAYS]
                                 Model IDs in your code: prices, retirement dates, cheaper options
  veqen diff [base-ref] [--format text|markdown|json]
                                 Cost impact of model changes (uses production volume with VEQEN_API_KEY)
  veqen audit [--provider openai|anthropic]
                                 Last 30 days of provider spend from an admin key, locally
  veqen report [ledger-file]     Costs by customer, feature and model from a local ledger
```

--------------------------------

### Install veqen Haystack integration and disable telemetry

Source: https://github.com/keyveve/veqen-sdks/blob/main/examples/haystack/README.md

Install the optional Haystack integration with pip and disable Haystack telemetry. Requires Python 3.11+ and Haystack 3.4+.

```shell
pip install 'veqen[haystack]>=1.2,<2'
export HAYSTACK_TELEMETRY_ENABLED=false
```

--------------------------------

### init

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/python-core.md

Configures the Veqen client. Without an API key, costs are computed locally and nothing is sent anywhere. Reads environment variables when arguments are omitted.

```APIDOC
## init

### Description
Configures the Veqen client. Without an API key, costs are computed locally (and written to `ledger_file` if set) and nothing is sent anywhere. Reads `VEQEN_API_KEY`, `VEQEN_ENDPOINT`, `VEQEN_ENVIRONMENT` and `VEQEN_LEDGER_FILE` when arguments are omitted.

### Method
Function call

### Signature
```python
def init(
    api_key: str | None = None,
    *,
    endpoint: str | None = None,
    environment: str | None = None,
    ledger_file: str | None = None,
    flush_interval: float = 2.0,
    budget_poll: float = 30.0,
) -> None:
```

### Parameters
- `api_key` (str | None) - Optional - When set, events are uploaded to the ledger endpoint and the background sender thread starts. Defaults to env `VEQEN_API_KEY` or `""`.
- `endpoint` (str | None) - Optional - Base URL, trailing `/` stripped. Defaults to env `VEQEN_ENDPOINT` or `"https://veqen-app.com"`.
- `environment` (str | None) - Optional - Default environment tag, applied to every event. Defaults to env `VEQEN_ENVIRONMENT` or `None`.
- `ledger_file` (str | None) - Optional - Append each event as one JSON line; parent folder is created (creation failure only logs a warning). Defaults to env `VEQEN_LEDGER_FILE` or `None`.
- `flush_interval` (float) - Optional - Seconds between sender wake-ups / batch attempts. Defaults to `2.0`.
- `budget_poll` (float) - Optional - Seconds between `GET /v1/budgets/status` polls. Defaults to `30.0`.

### Returns
`None`.

### Notes
Setting a non-empty `api_key` starts a daemon thread named `veqen-sender` and registers `atexit.register(self.flush, 2.0)`. Calling `init` again re-configures the singleton.

### Example
```python
import veqen

veqen.init(ledger_file=".veqen/ledger.jsonl", environment="production")
```

```python
import os
import veqen

veqen.init(
    api_key=os.environ["VEQEN_API_KEY"],
    endpoint=os.environ.get("VEQEN_ENDPOINT", "https://veqen-app.com"),
    flush_interval=1.0,   # batch faster in a serverless hot path
    budget_poll=15.0,     # poll dashboard budgets more often
)
```
```

--------------------------------

### cost usage example with unknown model

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/js-pricing.md

Example of calling cost with an unknown model, which returns null. The code checks for null and logs a message suggesting to pass costUsd to record().

```javascript
import { cost } from "@veqen/sdk";

const price = cost("unknown-model", { inputTokens: 10 });
if (price == null) console.log("not in the price book — pass costUsd to record()");
```

--------------------------------

### priceCall function and usage example

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/js-pricing.md

Calculates USD cost of a call, rounded to 10 decimals, or null when rates are missing. The TypeScript declaration and JavaScript usage example are shown together.

```javascript
/** USD cost of a call to `price`, or null when the call used something the model has no rate for. */
export function priceCall(price, usage, { batch = false, serviceTier, residency, variant } = {}) {
  ...
  return Math.round(total * factor * 1e10) / 1e10;
}
```

```typescript
export declare function priceCall(price: Price, usage: Usage, options?: CostOptions): number | null;
```

```javascript
import { lookup, priceCall } from "@veqen/sdk";

const price = lookup("gpt-4o");
const usd = priceCall(price, { inputTokens: 1000, outputTokens: 500 });
const batchUsd = priceCall(price, { inputTokens: 1000, outputTokens: 500 }, { batch: true });
```

--------------------------------

### VeqenTracingProcessor.on_span_start

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/python-integrations.md

Called when a span starts. This implementation is a no-op.

```APIDOC
## `VeqenTracingProcessor.on_span_start`

### Description
Called when a span starts. This implementation is a no-op.

### Method
Method

### Signature
```python
on_span_start(self, span: Any) -> None
```

### Parameters
- **span** (Any) - Required - The span object that started.

```

--------------------------------

### main

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/js-cli.md

Entry point for the veqen CLI. Parses command-line arguments and dispatches to subcommands (init, scan, diff, audit, report). Returns a Promise that resolves when the command completes.

```APIDOC
## main

### Description
Entry point for the veqen CLI. Parses command-line arguments and dispatches to subcommands (init, scan, diff, audit, report). Returns a Promise that resolves when the command completes.

### Method
CLI

### Endpoint
`veqen` (bin)

### Parameters
#### Positional Arguments
- **command** (string) - Required - The subcommand to execute: `init`, `scan`, `diff`, `audit`, or `report`.
- **arg** (string) - Optional - Argument for the subcommand (e.g., path, git ref, or ledger file).

#### Options
- **--json** (boolean) - Optional - Output in JSON format.
- **--format** (string) - Optional - Output format.
- **--provider** (string) - Optional - Provider name.
- **--fail-within** (string) - Optional - Fail threshold.
- **--help, -h** (boolean) - Optional - Show help.

### Returns
`Promise<void>`

### Errors
- Any command error prints `veqen: <message>` to stderr and sets `process.exitCode = 1`.
- Unknown/no command prints the help block.
```

--------------------------------

### Price a call without recording with @veqen/sdk

Source: https://github.com/keyveve/veqen-sdks/blob/main/js/README.md

Get the price for a model without recording usage. The price book is @veqen/sdk/prices.

```ts
cost("gpt-4o", { inputTokens: 1000, outputTokens: 500 })
```

--------------------------------

### main() CLI entry point

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/js-cli.md

Parses CLI arguments and dispatches to subcommands (init, scan, diff, audit, report). On error, prints `veqen: <message>` to stderr and sets `process.exitCode = 1`; unknown/no command prints the help block.

```JavaScript
export async function main(argv = process.argv.slice(2)) {
  const options = { json: { type: "boolean" }, format: { type: "string" }, provider: { type: "string" }, "fail-within": { type: "string" }, help: { type: "boolean", short: "h" } };
  try {
    const { positionals, values } = parseArgs({ args: argv, options, allowPositionals: true });
    const [command, arg] = values.help ? [] : positionals;
    if (command === "init") runInit(arg ?? ".");
    else if (command === "scan") runScan(arg ?? ".", values);
    else if (command === "diff") await runDiff(arg ?? "origin/main", values);
    else if (command === "audit") await runAudit(values);
    else if (command === "report") runReport(arg ?? process.env.VEQEN_LEDGER_FILE ?? ".veqen/ledger.jsonl");
    else console.log(HELP);
  } catch (error) {
    console.error(`veqen: ${error.message}`);
    process.exitCode = 1;
  }
}
```

--------------------------------

### init(options)

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/types.md

Initializes the Veqen SDK client with configuration options. Accepts an optional options object to set API key, endpoint, environment, ledger file, and polling intervals.

```APIDOC
## `init(options)`

### Description
Initializes the Veqen SDK client with configuration options. Accepts an optional options object to set API key, endpoint, environment, ledger file, and polling intervals.

### Method
`init`

### Parameters
#### Options
- **apiKey** (string) - Optional - API key for authentication. Defaults to env `VEQEN_API_KEY` or `""`.
- **endpoint** (string) - Optional - Base endpoint URL. Defaults to env `VEQEN_ENDPOINT` or `"https://veqen-app.com"`.
- **environment** (string) - Optional - Environment name. Defaults to env `VEQEN_ENVIRONMENT`.
- **ledgerFile** (string) - Optional - Path to ledger file. Defaults to env `VEQEN_LEDGER_FILE`.
- **flushIntervalMs** (number) - Optional - Flush interval in milliseconds. Defaults to `2000`.
- **budgetPollMs** (number) - Optional - Budget polling interval in milliseconds. Defaults to `30000`.

### Request Example
```ts
init({
  apiKey: "your-api-key",
  endpoint: "https://veqen-app.com",
  environment: "production",
  ledgerFile: "./ledger.json",
  flushIntervalMs: 2000,
  budgetPollMs: 30000
});
```

### Response
No return value.
```

--------------------------------

### GET https://veqen-app.com/prices.json

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/endpoints.md

Serves the deployed price book snapshot (schema 2). This is a public endpoint for retrieving the current price book data.

```APIDOC
## GET https://veqen-app.com/prices.json

### Description
Serves the deployed price book snapshot (schema 2). This is a public endpoint for retrieving the current price book data.

### Method
GET

### Endpoint
https://veqen-app.com/prices.json

### Response
#### Success Response (200)
- **schema** (integer) - Schema version, set to 2.
- **prices** (object) - Price book data.

#### Response Example
```json
{
  "schema": 2,
  "prices": {}
}
```
```

--------------------------------

### Catching BudgetExceeded

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/python-core.md

Example of catching BudgetExceeded when wrapping an OpenAI client with a budget. The exception provides scope, value, and limit_usd for handling.

```python
import veqen

veqen.budget(2.00, user="*")
try:
    client = veqen.wrap(OpenAI(), feature="chat")
    client.chat.completions.create(model="gpt-4o-mini", messages=[...])
except veqen.BudgetExceeded as exc:
    return upgrade_prompt(exc.limit_usd)  # exc.scope == "user", exc.value == "u_42"
```

--------------------------------

### Budget enforcement example

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/python-integrations.md

Sets a budget and catches BudgetExceeded to handle budget limits. The handler's raise_error attribute lets the exception propagate.

```python
import veqen
from veqen.integrations.langchain import VeqenCallbackHandler

handler = VeqenCallbackHandler(feature="agent")
veqen.budget(2.00, user="*")
try:
    chain.invoke(inputs, config={"callbacks": [handler]})
except veqen.BudgetExceeded:
    print("budget reached for this user")
```

--------------------------------

### veqen report .veqen/ledger.jsonl

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/js-cli.md

Reads the JSONL ledger file and prints the top 15 rows per dimension (customer, feature, model) with calls and cost. Throws an error if the ledger file is missing.

```bash
npx @veqen/sdk@1 report .veqen/ledger.jsonl
```

--------------------------------

### GET https://api.anthropic.com/v1/organizations/cost_report

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/endpoints.md

Retrieves cost report data from Anthropic's organization cost report endpoint. Used by the `veqen audit` command to fetch model-level costs.

```APIDOC
## GET https://api.anthropic.com/v1/organizations/cost_report

### Description
Retrieves cost report data from Anthropic's organization cost report endpoint. Used by the `veqen audit` command to fetch model-level costs.

### Method
GET

### Endpoint
https://api.anthropic.com/v1/organizations/cost_report

### Parameters
#### Query Parameters
- **starting_at** (string) - Required - Start date in `YYYY-MM-DDT00:00:00Z` format.
- **group_by[]** (string) - Required - Set to `description` to group by model description.
- **limit** (integer) - Required - Maximum number of results, set to `31`.
- **page** (integer) - Required - Page number for pagination.

### Response
#### Success Response (200)
- **data[].results[].model** (string) - Model identifier or description.
- **data[].results[].description** (string) - Description of the cost item.
- **data[].results[].amount** (number) - Cost amount in cents (convert to USD).
- **has_more** (boolean) - Indicates if more pages are available.
- **next_page** (string) - Token for the next page.

#### Response Example
```json
{
  "data": [
    {
      "results": [
        {
          "model": "claude-3-opus",
          "description": "Claude 3 Opus",
          "amount": 1234
        }
      ]
    }
  ],
  "has_more": false,
  "next_page": null
}
```
```

--------------------------------

### Local-only init with ledger file

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/python-core.md

Initializes Veqen without an API key, so costs are computed locally and appended to the specified ledger file. Sets a default environment tag for all events.

```python
import veqen

veqen.init(ledger_file=".veqen/ledger.jsonl", environment="production")
```

--------------------------------

### GET https://api.openai.com/v1/organization/costs

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/endpoints.md

Retrieves cost data from OpenAI's organization costs endpoint. Used by the `veqen audit` command to fetch line-item costs for the last 30 days.

```APIDOC
## GET https://api.openai.com/v1/organization/costs

### Description
Retrieves cost data from OpenAI's organization costs endpoint. Used by the `veqen audit` command to fetch line-item costs for the last 30 days.

### Method
GET

### Endpoint
https://api.openai.com/v1/organization/costs

### Parameters
#### Query Parameters
- **start_time** (unix timestamp) - Required - Start time for cost data, typically 30 days ago.
- **bucket_width** (string) - Required - Set to `1d` for daily buckets.
- **group_by** (string) - Required - Set to `line_item` to group costs by line item.
- **limit** (integer) - Required - Maximum number of results, set to `31`.
- **page** (integer) - Required - Page number for pagination.

### Response
#### Success Response (200)
- **data[].results[].line_item** (string) - Line item identifier, split on commas.
- **data[].results[].amount.value** (number) - Cost amount value.
- **has_more** (boolean) - Indicates if more pages are available.
- **next_page** (string) - Token for the next page.

#### Response Example
```json
{
  "data": [
    {
      "results": [
        {
          "line_item": "item1,item2",
          "amount": { "value": 12.34 }
        }
      ]
    }
  ],
  "has_more": false,
  "next_page": null
}
```
```

--------------------------------

### Using veqen.tags with a client call

Source: https://github.com/keyveve/veqen-sdks/blob/main/_autodocs/api-reference/python-core.md

Shows how to use the `tags` context manager to attribute a client call with tags like `customer` and `feature`.

```python
import veqen

with veqen.tags(customer="acme", feature="support-bot"):
    client.responses.create(model="gpt-5-mini", input="Where is my order?")
```

--------------------------------

### Load price book in JavaScript

Source: https://github.com/keyveve/veqen-sdks/blob/main/pricebook-packages/js/README.md

Load the price book and log version, schema, and rates for a model. Token rates are USD per million tokens.

```javascript
import book from '@veqen/pricebook';

console.log(book.version, book.schema);
console.log(book.models['openai/gpt-4o'].rates);
```
Tokens:
Raw