# Genkit Plugins
This directory contains all official Genkit plugins for Python.
## Plugin Architecture
All plugins inherit from `genkit.core.plugin.Plugin` and implement three
async methods. The registry calls them lazily — `init()` runs only on
first use, not at registration time.
```
┌─────────────────────────────────────────────────────────────────────┐
│ Plugin (Abstract Base Class) │
│ genkit.core.plugin.Plugin │
├─────────────────────────────────────────────────────────────────────┤
│ │
│ name: str Plugin namespace (e.g., 'googleai') │
│ │
│ async init() One-time setup; returns actions to │
│ → list[Action] pre-register. Called lazily on first use. │
│ │
│ async resolve(kind, name) Resolve a single action by kind/name. │
│ → Action | None Called on each action lookup. │
│ │
│ async list_actions() Advertise available actions (for Dev UI). │
│ → list[ActionMetadata] Must be fast — no heavy initialization.│
│ │
└─────────────────────────────────────────────────────────────────────┘
│
┌───────────┬───────────┼───────────┬───────────┐
▼ ▼ ▼ ▼ ▼
GoogleAI Anthropic Ollama Cohere ... etc.
```
### Plugin Lifecycle
```
ai = Genkit(plugins=[GoogleAI()]) ← Phase 1: REGISTER
│
▼
registry.register_plugin(GoogleAI()) (stored, not initialized)
│
⋮ (later, on first use)
│
await ai.generate(model="googleai/gemini-2.0-flash", ...)
│
▼
registry._ensure_plugin_initialized() ← Phase 2: LAZY INIT
│
▼
actions = await plugin.init() (called exactly once)
for action in actions:
registry.register_action_instance(action)
│
▼
await plugin.resolve(MODEL, name) ← Phase 3: RESOLVE
│
▼
Action cached in registry (subsequent lookups skip init)
```
### Action Resolution Algorithm
```
ai.generate(model="googleai/gemini-2.0-flash")
│
▼
┌──────────────────────────────────────────────────────────────────┐
│ 1. CACHE HIT? │
│ Name already in registry._entries? │
│ YES → return cached Action │ NO → continue │
├──────────────────────────────────────────────────────────────────┤
│ 2. NAMESPACED? (name contains "/") │
│ YES → find plugin by prefix, init it, resolve via plugin │
│ NO → try ALL plugins; 1 match = ok, 2+ = ValueError │
├──────────────────────────────────────────────────────────────────┤
│ 3. DYNAMIC ACTION PROVIDERS (fallback) │
│ Try MCP servers / dynamic providers → found? register it │
│ Not found → return None │
└──────────────────────────────────────────────────────────────────┘
```
## Plugin Categories
```
┌─────────────────────────────────────────────────────────────────────────────────┐
│ GENKIT PLUGIN ECOSYSTEM │
├─────────────────────────────────────────────────────────────────────────────────┤
│ │
│ MODEL PROVIDERS TELEMETRY │
│ ─────────────── ───────── │
│ ┌─────────────────────────┐ ┌─────────────────────────┐ │
│ │ google-genai │ │ google-cloud │ │
│ │ • Gemini, Imagen, Veo │ │ • Cloud Trace │ │
│ │ • Lyria, TTS │ │ • Cloud Logging │ │
│ └─────────────────────────┘ └─────────────────────────┘ │
│ ┌─────────────────────────┐ ┌─────────────────────────┐ │
│ │ anthropic │ │ firebase │ │
│ │ • Claude 3.5/4 │ │ • Firebase Telemetry │ │
│ └─────────────────────────┘ └─────────────────────────┘ │
│ ┌─────────────────────────┐ │
│ │ amazon-bedrock 🌐 │ INTEGRATIONS │
│ │ • Claude, Llama, Nova │ ──────────── │
│ │ • Titan, Mistral │ ┌─────────────────────────┐ │
│ │ • X-Ray telemetry │ │ flask │ │
│ └─────────────────────────┘ │ • HTTP endpoints │ │
│ ┌─────────────────────────┐ └─────────────────────────┘ │
│ │ microsoft-foundry │ ┌─────────────────────────┐ │
│ │ • GPT-4o, Claude, Llama │ │ mcp │ │
│ │ • 11,000+ models │ │ • Model Context Protocol│ │
│ └─────────────────────────┘ └─────────────────────────┘ │
│ ┌─────────────────────────┐ │
│ │ vertex-ai │ VECTOR STORES │
│ │ • Model Garden │ ───────────── │
│ │ • Vector Search │ ┌─────────────────────────┐ │
│ └─────────────────────────┘ │ firebase │ │
│ ┌─────────────────────────┐ │ • Firestore vectors │ │
│ │ ollama │ └─────────────────────────┘ │
│ │ • Local models │ ┌─────────────────────────┐ │
│ └─────────────────────────┘ │ vertex-ai │ │
│ ┌─────────────────────────┐ │ • Vector Search │ │
│ │ compat-oai │ └─────────────────────────┘ │
│ │ • OpenAI API compatible │ ┌─────────────────────────┐ │
│ └─────────────────────────┘ │ dev-local-vectorstore │ │
│ ┌─────────────────────────┐ │ • Local development │ │
│ │ deepseek │ └─────────────────────────┘ │
│ │ • DeepSeek V3, R1 │ │
│ └─────────────────────────┘ SAFETY & EVALUATION │
│ ┌─────────────────────────┐ ─────────────────── │
│ │ xai │ ┌─────────────────────────┐ │
│ │ • Grok models │ │ checks │ │
│ └─────────────────────────┘ │ • Content moderation │ │
│ ┌─────────────────────────┐ │ • Safety guardrails │ │
│ │ mistral │ ┌─────────────────────────┐ │
│ │ • Mistral Large, Small │ │ evaluators │ │
│ │ • Codestral, Pixtral │ │ • RAGAS metrics │ │
│ └─────────────────────────┘ │ • Custom evaluators │ │
│ ┌─────────────────────────┐ └─────────────────────────┘ │
│ │ huggingface │ │
│ │ • 1M+ open models │ │
│ │ • Inference providers │ │
│ └─────────────────────────┘ │
│ ┌─────────────────────────┐ │
│ │ cohere 🌐 │ │
│ │ • Command R/R+ │ │
│ │ • Embed, Rerank │ │
│ └─────────────────────────┘ │
│ │
└─────────────────────────────────────────────────────────────────────────────────┘
```
## When to Use What
### Model Provider Selection
```
┌─────────────────────────────────────────────────────────────────────────────────┐
│ WHICH MODEL PROVIDER SHOULD I USE? │
├─────────────────────────────────────────────────────────────────────────────────┤
│ │
│ "I want the best multimodal AI" │
│ → google-genai (Gemini 2.0) │
│ │
│ "I need Claude models" │
│ → anthropic (direct) OR amazon-bedrock OR microsoft-foundry │
│ │
│ "I'm on AWS and want managed models" │
│ → amazon-bedrock (Claude, Llama, Nova, Titan) │
│ │
│ "I'm on Azure and want managed models" │
│ → microsoft-foundry (GPT-4o, Claude, Llama, 11,000+ models) │
│ │
│ "I'm on GCP and want third-party models" │
│ → vertex-ai (Model Garden - Claude, Llama, etc.) │
│ │
│ "I want to run models locally" │
│ → ollama (Llama, Mistral, Phi, etc.) │
│ │
│ "I need OpenAI GPT models" │
│ → compat-oai (direct OpenAI) OR microsoft-foundry (via Azure) │
│ │
│ "I want to use any OpenAI-compatible API" │
│ → compat-oai (works with OpenRouter, Together, etc.) │
│ │
│ "I need DeepSeek reasoning models" │
│ → deepseek (V3, R1 reasoning) │
│ │
│ "I want Grok models" │
│ → xai │
│ │
│ "I want Cloudflare Workers AI models" │
│ → cloudflare-workers-ai (Llama, Mistral, Qwen + OTLP telemetry) │
│ │
│ "I want Mistral AI models (French AI)" │
│ → mistral (mistral-large, codestral, pixtral) │
│ │
│ "I want access to 1M+ open source models" │
│ → huggingface (Inference API + 17 providers) │
│ │
│ "I want Cohere command models + reranking" │
│ → cohere (Command R/R+, Embed, Rerank) │
│ │
│ "I want one API for 500+ models from 60+ providers" │
│ → compat-oai with OpenRouter (works TODAY) │
│ OR openrouter plugin (COMING SOON - adds model discovery) │
│ │
└─────────────────────────────────────────────────────────────────────────────────┘
```
### Telemetry Selection
```
┌─────────────────────────────────────────────────────────────────────────────────┐
│ WHICH TELEMETRY PLUGIN SHOULD I USE? │
├─────────────────────────────────────────────────────────────────────────────────┤
│ │
│ NATIVE PLATFORM BACKENDS THIRD-PARTY BACKENDS │
│ ──────────────────────── ──────────────────── │
│ │
│ ┌─────────┐ ┌─────────┐ ┌───────────────────┐ │
│ │ amazon- │ │ google- │ │ observability │ │
│ │ bedrock │ │ cloud │ │ • Sentry │ │
│ │ • SigV4 │ │ • ADC │ │ • Honeycomb │ │
│ │ • X-Ray │ │ • Trace │ │ • Datadog │ │
│ │ │ │ • Logs │ │ • Grafana │ │
│ └────┬────┘ └────┬────┘ │ • Axiom │ │
│ │ │ └─────────┬─────────┘ │
│ ▼ ▼ │ │
│ ┌─────────┐ ┌─────────┐ ▼ │
│ │ X-Ray │ │ Cloud │ ┌───────────────────┐ │
│ │ Console │ │ Trace │ │ Any OTLP Backend │ │
│ └─────────┘ └─────────┘ └───────────────────┘ │
│ │
│ ┌───────────────────┐ ┌─────────────┐ │
│ │ microsoft-foundry │ │cloudflare-workers-ai│ │
│ │ • Models + AppIns │ │ • OTLP │ • Models + Telemetry │
│ │ • Azure Telemetry │ │ • Token │ • Single plugin │
│ └─────────┬─────────┘ └──────┬──────┘ │
│ │ │ CAN'T BE REPLICATED CAN BE REPLICATED │
│ ▼ ▼ WITH GENERIC OTLP WITH GENERIC OTLP │
│ ┌───────────────────┐ ┌─────────┐ │
│ │ App Insights │ │ OTLP │ │
│ └───────────────────┘ │ Backend │ │
│ └─────────┘ │
│ │
├─────────────────────────────────────────────────────────────────────────────────┤
│ │
│ "I'm on AWS and want X-Ray" → amazon-bedrock plugin │
│ "I'm on GCP and want Cloud Trace" → google-cloud plugin │
│ "I'm on Azure and want App Insights" → microsoft-foundry plugin │
│ "I'm using Firebase" → firebase plugin (auto telemetry) │
│ │
│ "I want Sentry/Honeycomb/Datadog" → observability plugin │
│ "I'm multi-cloud" → observability plugin │
│ "I want generic OTLP export" → cloudflare-workers-ai plugin (combined) │
│ │
└─────────────────────────────────────────────────────────────────────────────────┘
```
### Vector Store Selection
```
┌─────────────────────────────────────────────────────────────────────────────────┐
│ WHICH VECTOR STORE SHOULD I USE? │
├─────────────────────────────────────────────────────────────────────────────────┤
│ │
│ "I'm just developing/testing locally" │
│ → dev-local-vectorstore │
│ │
│ "I need production vector search on Firebase" │
│ → firebase (Firestore vector search) │
│ │
│ "I need enterprise-scale vector search on GCP" │
│ → vertex-ai (Vertex AI Vector Search + Firestore/BigQuery) │
│ │
│ "I want to use a third-party vector DB" │
│ → Implement custom retriever (Pinecone, Weaviate, Chroma, etc.) │
│ │
└─────────────────────────────────────────────────────────────────────────────────┘
```
### Safety & Evaluation Selection
```
┌─────────────────────────────────────────────────────────────────────────────────┐
│ WHICH SAFETY/EVALUATION PLUGIN SHOULD I USE? │
├─────────────────────────────────────────────────────────────────────────────────┤
│ │
│ "I need content moderation / safety guardrails" │
│ → checks (Google Checks AI Safety) │
│ • Real-time content classification │
│ • Block harmful input/output │
│ • 8 policy types (harassment, hate speech, etc.) │
│ │
│ "I need to evaluate RAG quality" │
│ → evaluators (RAGAS metrics) │
│ • Faithfulness, relevancy, answer accuracy │
│ • Custom evaluation metrics │
│ │
│ "I need both safety AND quality evaluation" │
│ → Use both: checks for guardrails, evaluators for quality │
│ │
└─────────────────────────────────────────────────────────────────────────────────┘
```
## Plugin List
### Model Providers
| Plugin | Models | Best For |
|--------|--------|----------|
| **google-genai** | Gemini, Imagen, Veo, Lyria | Multimodal AI, Google ecosystem |
| **anthropic** | Claude 3.5, Claude 4 | Direct Claude access |
| **amazon-bedrock** 🌐 | Claude, Llama, Nova, Titan | AWS managed models (community) |
| **microsoft-foundry** 🌐 | GPT-4o, Claude, Llama, 11,000+ | Azure AI, enterprise (community) |
| **vertex-ai** | Model Garden (Claude, Llama) | GCP third-party models |
| **ollama** | Llama, Mistral, Phi, etc. | Local/private deployment |
| **compat-oai** | Any OpenAI-compatible | OpenAI, OpenRouter, etc. |
| **deepseek** | DeepSeek V3, R1 | Reasoning, cost-effective |
| **xai** | Grok | X/Twitter integration |
| **cloudflare-workers-ai** 🌐 | Llama, Mistral, Qwen, Gemma | Cloudflare Workers AI + OTLP telemetry (community) |
| **mistral** | Mistral Large, Small, Codestral, Pixtral | French AI, efficient models, code generation |
| **huggingface** | 1M+ models via HF Hub | Open source models, inference providers |
| **cohere** 🌐 | Command R/R+, Embed, Rerank | Cohere models, embeddings, reranking (community) |
### Planned Model Providers
| Plugin | Models | Status | Notes |
|--------|--------|--------|-------|
| **openrouter** | 500+ models, 60+ providers | 🔜 Planned | Unified gateway (OpenAI, Anthropic, Google, etc.) |
> **Note:** OpenRouter is already usable today via `compat-oai` since it's OpenAI-compatible.
> A dedicated plugin would add model discovery, provider routing, and usage analytics.
### Telemetry
| Plugin | Backend | Features |
|--------|---------|----------|
| **google-cloud** | Cloud Trace, Logging | GCP native, log correlation |
| **amazon-bedrock** 🌐 | X-Ray | AWS native, SigV4 auth, built into model plugin (community) |
| **microsoft-foundry** 🌐 | Application Insights | Azure Monitor, trace correlation, built into model plugin (community) |
| **cloudflare-workers-ai** 🌐 | Any OTLP endpoint | Generic OTLP, Bearer auth, combined with models (community) |
| **observability** 🌐 | Sentry, Honeycomb, Datadog, Grafana, Axiom | 3rd party presets (community) |
| **firebase** | Firebase console | Auto-telemetry for Firebase apps |
### Integrations
| Plugin | Purpose |
|--------|---------|
| **flask** | Serve flows via Flask HTTP endpoints |
| **mcp** | Model Context Protocol for tool integration |
### Safety & Evaluation
| Plugin | Purpose | Features |
|--------|---------|----------|
| **checks** | Content moderation & safety | Google Checks AI guardrails, 8 policy types |
| **evaluators** | Quality evaluation | RAGAS metrics, custom evaluators |
### Vector Stores
| Plugin | Backend | Scale |
|--------|---------|-------|
| **dev-local-vectorstore** | Local JSON | Development only |
| **firebase** | Firestore | Production, serverless |
| **vertex-ai** | Vertex AI Vector Search | Enterprise scale |
## Environment Variables Reference
All environment variables used by Genkit plugins. Configure these before running your application.
### Model Provider Environment Variables
| Variable | Plugin | Required | Description | Documentation |
|----------|--------|----------|-------------|---------------|
| `GOOGLE_GENAI_API_KEY` | google-genai | Yes | Google AI Studio API key | [Get API Key](https://aistudio.google.com/apikey) |
| `ANTHROPIC_API_KEY` | anthropic | Yes | Anthropic API key | [Anthropic Console](https://console.anthropic.com/) |
| `AWS_REGION` | amazon-bedrock | Yes | AWS region (e.g., `us-east-1`) | [AWS Regions](https://docs.aws.amazon.com/general/latest/gr/bedrock.html) |
| `AWS_ACCESS_KEY_ID` | amazon-bedrock | Yes* | AWS access key | [AWS Credentials](https://docs.aws.amazon.com/cli/latest/userguide/cli-configure-envvars.html) |
| `AWS_SECRET_ACCESS_KEY` | amazon-bedrock | Yes* | AWS secret key | [AWS Credentials](https://docs.aws.amazon.com/cli/latest/userguide/cli-configure-envvars.html) |
| `AZURE_AI_FOUNDRY_ENDPOINT` | microsoft-foundry | Yes | Azure AI Foundry endpoint URL | [Azure AI Foundry](https://ai.azure.com/) |
| `AZURE_AI_FOUNDRY_API_KEY` | microsoft-foundry | Yes* | Azure AI Foundry API key | [Azure AI Foundry](https://ai.azure.com/) |
| `OPENAI_API_KEY` | compat-oai | Yes | OpenAI API key | [OpenAI API Keys](https://platform.openai.com/api-keys) |
| `OPENAI_ORG_ID` | compat-oai | No | OpenAI organization ID | [OpenAI Settings](https://platform.openai.com/account/organization) |
| `DEEPSEEK_API_KEY` | deepseek | Yes | DeepSeek API key | [DeepSeek Platform](https://platform.deepseek.com/) |
| `XAI_API_KEY` | xai | Yes | xAI API key | [xAI Console](https://console.x.ai/) |
| `CLOUDFLARE_ACCOUNT_ID` | cloudflare-workers-ai | Yes | Cloudflare account ID | [Cloudflare Dashboard](https://dash.cloudflare.com/) |
| `CLOUDFLARE_API_TOKEN` | cloudflare-workers-ai | Yes | Cloudflare API token | [Cloudflare API Tokens](https://developers.cloudflare.com/fundamentals/api/get-started/create-token/) |
| `MISTRAL_API_KEY` | mistral | Yes | Mistral AI API key | [Mistral Console](https://console.mistral.ai/) |
| `HF_TOKEN` | huggingface | Yes | Hugging Face API token | [HF Tokens](https://huggingface.co/settings/tokens) |
| `COHERE_API_KEY` | cohere | Yes | Cohere API key | [Cohere Dashboard](https://dashboard.cohere.com/api-keys) |
| `CO_API_KEY` | cohere | Yes* | Cohere API key (alternative) | [Cohere Dashboard](https://dashboard.cohere.com/api-keys) |
*Can use IAM roles, managed identity, or other credential providers instead. For cohere, either `COHERE_API_KEY` or `CO_API_KEY` is accepted.
### Telemetry Environment Variables
#### Google Cloud Plugin
| Variable | Required | Description | Documentation |
|----------|----------|-------------|---------------|
| `GOOGLE_CLOUD_PROJECT` | Yes | GCP project ID | [GCP Projects](https://cloud.google.com/resource-manager/docs/creating-managing-projects) |
| `GOOGLE_APPLICATION_CREDENTIALS` | Yes* | Path to service account JSON | [Application Default Credentials](https://cloud.google.com/docs/authentication/application-default-credentials) |
| `GCLOUD_PROJECT` | No | Alternative to `GOOGLE_CLOUD_PROJECT` | - |
| `FIREBASE_PROJECT_ID` | No | Firebase project ID (auto-detected) | - |
*Not required when running on GCP with default credentials.
#### Amazon Bedrock Plugin (X-Ray Telemetry)
| Variable | Required | Description | Documentation |
|----------|----------|-------------|---------------|
| `AWS_REGION` | Yes | AWS region for X-Ray | [AWS X-Ray](https://docs.aws.amazon.com/xray/latest/devguide/xray-sdk-python.html) |
| `AWS_ACCESS_KEY_ID` | Yes* | AWS access key | [AWS Credentials](https://docs.aws.amazon.com/cli/latest/userguide/cli-configure-envvars.html) |
| `AWS_SECRET_ACCESS_KEY` | Yes* | AWS secret key | [AWS Credentials](https://docs.aws.amazon.com/cli/latest/userguide/cli-configure-envvars.html) |
| `AWS_DEFAULT_REGION` | No | Fallback region | - |
*Can use IAM roles instead.
#### Microsoft Foundry Plugin (Azure Telemetry)
| Variable | Required | Description | Documentation |
|----------|----------|-------------|---------------|
| `APPLICATIONINSIGHTS_CONNECTION_STRING` | Yes | Application Insights connection string | [Azure Monitor OpenTelemetry](https://learn.microsoft.com/azure/azure-monitor/app/opentelemetry-configuration) |
| `AZURE_TENANT_ID` | No | Azure AD tenant ID | [Azure Identity](https://learn.microsoft.com/azure/developer/python/sdk/authentication-overview) |
| `AZURE_CLIENT_ID` | No | Azure AD client ID | - |
| `AZURE_CLIENT_SECRET` | No | Azure AD client secret | - |
#### Cloudflare Workers AI (cloudflare-workers-ai) Plugin
| Variable | Required | Description | Documentation |
|----------|----------|-------------|---------------|
| `CLOUDFLARE_ACCOUNT_ID` | Yes | Cloudflare account ID | [Cloudflare Dashboard](https://dash.cloudflare.com/) |
| `CLOUDFLARE_API_TOKEN` | Yes | API token for Workers AI | [Cloudflare API Tokens](https://developers.cloudflare.com/fundamentals/api/get-started/create-token/) |
| `CF_OTLP_ENDPOINT` | No* | OTLP endpoint URL (for telemetry) | [Cloudflare Workers Observability](https://developers.cloudflare.com/workers/observability/) |
| `CF_API_TOKEN` | No | API token for telemetry (Bearer auth) | [Cloudflare API Tokens](https://developers.cloudflare.com/fundamentals/api/get-started/create-token/) |
*Required only if using OTLP telemetry export.
#### Observability Plugin (Third-Party Backends)
##### Sentry
| Variable | Required | Description | Documentation |
|----------|----------|-------------|---------------|
| `SENTRY_DSN` | Yes | Sentry DSN (Data Source Name) | [Sentry DSN](https://docs.sentry.io/concepts/otlp/) |
| `SENTRY_ENVIRONMENT` | No | Environment name (production, staging) | [Sentry Configuration](https://docs.sentry.io/platforms/python/configuration/options/) |
| `SENTRY_RELEASE` | No | Release version identifier | - |
##### Honeycomb
| Variable | Required | Description | Documentation |
|----------|----------|-------------|---------------|
| `HONEYCOMB_API_KEY` | Yes | Honeycomb API key | [Honeycomb API Keys](https://docs.honeycomb.io/configure/environments/manage-api-keys/) |
| `HONEYCOMB_DATASET` | No | Dataset name (Classic only) | [Honeycomb Datasets](https://docs.honeycomb.io/send-data/python/opentelemetry-sdk/) |
| `HONEYCOMB_API_ENDPOINT` | No | API endpoint (default: US) | [Honeycomb Endpoints](https://docs.honeycomb.io/configure/environments/manage-api-keys/) |
Honeycomb endpoints:
- US (default): `https://api.honeycomb.io`
- EU: `https://api.eu1.honeycomb.io`
##### Datadog
| Variable | Required | Description | Documentation |
|----------|----------|-------------|---------------|
| `DD_API_KEY` | Yes | Datadog API key | [Datadog API Keys](https://docs.datadoghq.com/account_management/api-app-keys/) |
| `DD_SITE` | No | Datadog site (default: `datadoghq.com`) | [Datadog Sites](https://docs.datadoghq.com/getting_started/site/) |
| `DD_APP_KEY` | No | Datadog application key | - |
Datadog sites: `datadoghq.com`, `datadoghq.eu`, `us3.datadoghq.com`, `us5.datadoghq.com`, `ap1.datadoghq.com`
##### Grafana Cloud
| Variable | Required | Description | Documentation |
|----------|----------|-------------|---------------|
| `GRAFANA_OTLP_ENDPOINT` | Yes | Grafana Cloud OTLP endpoint | [Grafana Cloud OTLP](https://grafana.com/docs/grafana-cloud/monitor-applications/application-observability/setup/collector/opentelemetry-collector/) |
| `GRAFANA_USER_ID` | Yes | Grafana Cloud instance ID (numeric) | [Grafana Cloud Portal](https://grafana.com/docs/grafana-cloud/account-management/authentication-and-permissions/) |
| `GRAFANA_API_KEY` | Yes | Grafana Cloud API key | [Grafana Cloud API Keys](https://grafana.com/docs/grafana-cloud/account-management/authentication-and-permissions/create-api-key/) |
Find your credentials: My Account > [Stack] > OpenTelemetry > Configure
##### Axiom
| Variable | Required | Description | Documentation |
|----------|----------|-------------|---------------|
| `AXIOM_TOKEN` | Yes | Axiom API token | [Axiom API Tokens](https://axiom.co/docs/reference/tokens) |
| `AXIOM_DATASET` | No | Dataset name (default: `genkit`) | [Axiom Datasets](https://axiom.co/docs/reference/datasets) |
| `AXIOM_ORG_ID` | No | Organization ID | - |
#### Generic OpenTelemetry (Standard Variables)
| Variable | Required | Description | Documentation |
|----------|----------|-------------|---------------|
| `OTEL_EXPORTER_OTLP_ENDPOINT` | Yes | OTLP collector endpoint | [OTel SDK Environment Variables](https://opentelemetry.io/docs/specs/otel/configuration/sdk-environment-variables/) |
| `OTEL_EXPORTER_OTLP_HEADERS` | No | Headers for authentication | - |
| `OTEL_SERVICE_NAME` | No | Service name for traces | - |
### Safety & Evaluation Environment Variables
| Variable | Plugin | Required | Description | Documentation |
|----------|--------|----------|-------------|---------------|
| `GOOGLE_CLOUD_PROJECT` | checks | Yes | GCP project with Checks API enabled | [Google Checks](https://developers.google.com/checks) |
| `GOOGLE_APPLICATION_CREDENTIALS` | checks | Yes* | Service account credentials | - |
## Installation
Each plugin is a separate package. Install only what you need:
```bash
# Model providers
pip install genkit-google-genai-plugin
pip install genkit-anthropic-plugin
pip install genkit-amazon-bedrock-plugin # Also includes X-Ray telemetry
pip install genkit-microsoft-foundry-plugin
# Telemetry
pip install genkit-google-cloud-plugin
# Safety & Evaluation
pip install genkit-checks-plugin
pip install genkit-evaluators-plugin
# Integrations
pip install genkit-flask-plugin
pip install genkit-mcp-plugin
pip install genkit-cohere-plugin
```
## Quick Start
```python
from genkit.ai import Genkit
from genkit.plugins.google_genai import GoogleAI
# Initialize with your chosen plugin
ai = Genkit(
plugins=[GoogleAI()],
model="googleai/gemini-2.0-flash",
)
@ai.flow()
async def hello(name: str) -> str:
response = await ai.generate(prompt=f"Say hello to {name}")
return response.text
```
## Plugin Dependency Graph
Shows how plugins relate to each other and the core `genkit` package. Most
plugins are independent leaf nodes; only a few have inter-plugin dependencies.
```
┌─────────────────────────────────────────────────────────────────────────────────┐
│ PLUGIN DEPENDENCY GRAPH │
├─────────────────────────────────────────────────────────────────────────────────┤
│ │
│ ┌──────────┐ │
│ │ genkit │ (core SDK) │
│ └─────┬────┘ │
│ │ │
│ ┌──────────────────┼──────────────────┐ │
│ │ │ │ │
│ ▼ ▼ ▼ │
│ ┌──────────────────┐ ┌──────────────┐ ┌───────────────────┐ │
│ │ compat-oai │ │ google-genai │ │ All other plugins │ │
│ │ (OpenAI compat) │ │ │ │ (independent) │ │
│ └────────┬─────────┘ └──────┬───────┘ └───────────────────┘ │
│ │ │ │
│ ┌──────┴──────┐ │ │
│ │ │ │ │
│ ▼ ▼ ▼ │
│ ┌─────────┐ ┌──────────┐ ┌──────────┐ │
│ │deepseek │ │vertex-ai │ │ flask │ │
│ │(extends)│ │(Model │ │(uses │ │
│ │ │ │ Garden) │ │ google- │ │
│ │ │ │ │ │ genai) │ │
│ └─────────┘ └──────────┘ └──────────┘ │
│ │
│ INDEPENDENT PLUGINS (no inter-plugin dependencies): │
│ ───────────────────────────────────────────────── │
│ google-genai, anthropic, amazon-bedrock, microsoft-foundry, │
│ ollama, xai, mistral, huggingface, cloudflare-workers-ai, │
│ cohere, google-cloud, firebase, observability, mcp, evaluators, │
│ dev-local-vectorstore, checks │
│ │
└─────────────────────────────────────────────────────────────────────────────────┘
```
**Key relationships:**
- **`compat-oai`** provides the shared OpenAI-compatible model layer (chat, image, TTS, STT)
- **`deepseek`** extends `compat-oai` with reasoning model detection and param validation
- **`vertex-ai`** (Model Garden) uses `compat-oai` for third-party model support
- **`flask`** has a dev dependency on `google-genai` for its sample
## Cross-Language Plugin Coverage
> **Last audited**: 2026-02-08
The table below compares plugin availability across Python and JavaScript SDKs.
Python currently has **21 plugins** vs JavaScript's **17 plugins**, with broader
model provider diversity.
### Model Providers
| Plugin | Python | JavaScript | Notes |
|--------|:------:|:----------:|-------|
| Google GenAI (Gemini, Imagen, Veo, Lyria) | ✅ | ✅ | |
| Vertex AI (Model Garden, Vector Search) | ✅ | ✅ | |
| Anthropic (Claude) | ✅ | ✅ | |
| Ollama | ✅ | ✅ | |
| OpenAI-Compatible (compat-oai) | ✅ | ✅ | |
| Amazon Bedrock | ✅ | — | Python-only; community 🌐 |
| Microsoft Foundry (Azure AI) | ✅ | — | Python-only; community 🌐 |
| DeepSeek | ✅ | — | Python-only |
| xAI (Grok) | ✅ | — | Python-only |
| Cloudflare Workers AI | ✅ | — | Python-only; community 🌐 |
| Mistral | ✅ | — | Python-only |
| HuggingFace | ✅ | — | Python-only |
| Cohere | ✅ | — | Python-only; community 🌐 |
### Telemetry & Observability
| Plugin | Python | JavaScript | Notes |
|--------|:------:|:----------:|-------|
| Google Cloud (Trace, Logging) | ✅ | ✅ | |
| Firebase | ✅ | ✅ | |
| Observability (Sentry, Honeycomb, Datadog, Grafana, Axiom) | ✅ | — | Python-only; community 🌐 |
### Integrations
| Plugin | Python | JavaScript | Notes |
|--------|:------:|:----------:|-------|
| MCP (Model Context Protocol) | ✅ | ✅ | |
| Flask | ✅ | — | Python-only |
| Express | — | ✅ | JS-only |
| Next.js | — | ✅ | JS-only |
### Vector Stores
| Plugin | Python | JavaScript | Notes |
|--------|:------:|:----------:|-------|
| Dev Local Vectorstore | ✅ | ✅ | |
| Firebase (Firestore vectors) | ✅ | ✅ | |
| Vertex AI Vector Search | ✅ | ✅ | |
| Chroma | — | ✅ | JS-only |
| Pinecone | — | ✅ | JS-only |
| Cloud SQL PG | — | ✅ | JS-only |
### Safety & Evaluation
| Plugin | Python | JavaScript | Notes |
|--------|:------:|:----------:|-------|
| Evaluators (RAGAS) | ✅ | ✅ | |
| Checks (Content Safety) | ✅ | ✅ | |
### Other
| Plugin | Python | JavaScript | Notes |
|--------|:------:|:----------:|-------|
| LangChain | — | ✅ | JS-only |
## Further Reading
- [Plugin Planning & Roadmap](../engdoc/planning/)
- [Feature Matrix](../engdoc/planning/FEATURE_MATRIX.md)
- [Contributing Guide](../engdoc/contributing/)
