ai connectors¶
AI and machine-learning service integrations. All connectors in this category target hosted inference APIs (chat completion, embeddings, image generation) and use HTTP REST over net/http with a per-provider auth scheme (mostly Bearer tokens). They are stdlib-only — no vendor SDKs — so they stay light, but they trade off advanced features (e.g. native streaming back-pressure) compared to the dedicated pkg/component/anthropic/, pkg/component/openai/, etc. components that ship with Helix. Use the connectors when you want low-friction integration; use the components when you need streaming, token-usage headers, or tool-calling in the agent pipeline.
Connectors in this category¶
| Connector | URI / type | Auth | Native SDK | One-line description |
|---|---|---|---|---|
anthropic_connector |
connector://anthropic-conn/send |
API key (x-api-key + anthropic-version) |
No (HTTP) | Anthropic Claude /v1/messages chat completion. |
cohere |
connector://cohere-conn/send |
Bearer API key | No (HTTP) | Cohere generate/embed/rerank NLP API. |
deepseek |
connector://deepseek-conn/send |
Bearer API key | No (HTTP) | DeepSeek chat-completion API (OpenAI-compatible). |
fireworks |
connector://fireworks-conn/send |
Bearer API key | No (HTTP) | Fireworks AI hosted inference (Llama, Mixtral, custom models). |
gemini_connector |
connector://gemini-conn/send |
API key (query param) | No (HTTP) | Google Gemini generateContent REST endpoint. |
groq |
connector://groq-conn/send |
Bearer API key | No (HTTP) | Groq LPU-backed low-latency inference. |
huggingface |
connector://hf-conn/send |
Bearer HF token | No (HTTP) | Hugging Face Inference API for any hosted model. |
mistral |
connector://mistral-conn/send |
Bearer API key | No (HTTP) | Mistral AI chat-completion + embeddings. |
ollama_connector |
connector://ollama-conn/send |
None (local) | No (HTTP) | Local Ollama daemon at http://localhost:11434. |
openai_connector |
connector://openai-conn/send |
Bearer API key | No (HTTP) | OpenAI chat completions, embeddings, image generation. |
perplexity |
connector://perplexity-conn/send |
Bearer API key | No (HTTP) | Perplexity search-augmented chat (Sonar models). |
replicate |
connector://replicate-conn/send |
Bearer API token | No (HTTP) | Replicate model hosting (image / video / custom). |
stability |
connector://stability-conn/send |
Bearer API key | No (HTTP) | Stability AI text-to-image generation (SD3, SDXL). |
together |
connector://together-conn/send |
Bearer API key | No (HTTP) | Together AI hosted open-source model inference. |
The Native SDK column is "Yes" for connectors that use a real Go SDK rather than HTTP REST. None of the AI connectors qualify — see the dedicated openai://, anthropic://, ollama://, vertex://, bedrock://, cohere://, mistral://, groq://, fireworks://, together://, replicate://, huggingface://, xai://, deepinfra:// components for higher-fidelity LLM integration.
Typical config¶
connectors:
openai-conn:
type: openai_connector
config:
api_key: ${OPENAI_API_KEY}
model: gpt-4o
max_tokens: 1024
temperature: 1
timeout: 120
workflows:
- id: chat-then-slack
trigger:
uri: direct://user-message
steps:
- to: connector://openai-conn/send
- to: connector://slack-bot/send
See also¶
- Connector SDK guide — adding a new connector
- Related component schemes for production LLM use:
openai://,anthropic://,ollama://,bedrock://,vertex://,cohere://,groq://,mistral://,fireworks://,together://,replicate://,huggingface://,xai://,deepinfra:// - Agent framework: reference/agent/ for ReAct, supervisor, plan-execute, deep, sequential, parallel, and loop agents