AI Capabilities

Every task independently selects its AI provider and model. No global lock-in — mix OpenAI, Anthropic, Gemini, and Ollama in the same workflow.

Multi-Provider Chat

OpenAI

- name: summarize
  type: ai.openai.chat
  properties:
    model: gpt-4o-mini
    messages:
      - role: system
        content: You are a concise summarizer.
      - role: user
        content: 'Summarize: {{ outputs["fetch"]["body"] }}'
    temperature: 0.3
    max_tokens: 500

Outputs: text, model, prompt_tokens, completion_tokens, total_tokens

Anthropic (Claude)

- name: analyze
  type: ai.anthropic.chat
  properties:
    model: claude-sonnet-4-20250514
    messages:
      - role: user
        content: "Analyze this dataset and find anomalies."
    max_tokens: 1000

Google Gemini

- name: translate
  type: ai.gemini.chat
  properties:
    model: gemini-2.0-flash
    messages:
      - role: user
        content: "Translate to Japanese: Hello, how are you?"

Ollama (Local Models)

- name: local-inference
  type: ai.ollama.chat
  properties:
    base_url: http://ollama:11434
    model: llama3.2
    messages:
      - role: user
        content: "Explain this error log."

Run any model locally — no API keys, no data leaving your network.

Embeddings

Generate vector embeddings for RAG pipelines.

- name: embed-docs
  type: ai.openai.embedding
  properties:
    model: text-embedding-3-small
    input: '{{ outputs["fetch_docs"]["body"] }}'

Outputs: embedding (float array), dimensions, model

Image Generation

- name: generate-thumbnail
  type: ai.openai.image
  properties:
    prompt: "A minimalist logo for a workflow automation tool"
    size: "1024x1024"
    model: dall-e-3

MCP Tool Integration

Connect to any MCP-compatible tool server. The AI agent calls tools through the standard Model Context Protocol.

MCP via SSE (Remote Server)

- name: mcp-tools
  type: ai.openai.chat
  properties:
    model: gpt-4o
    messages:
      - role: user
        content: "Use the available tools to find recent GitHub issues."
    mcp:
      url: http://mcp-server:3001/sse

MCP via Docker

- name: github-agent
  type: ai.openai.chat
  properties:
    model: gpt-4o
    messages:
      - role: user
        content: "List open PRs in excalibase/workflow"
    mcp:
      image: mcp/github
      env:
        GITHUB_PERSONAL_ACCESS_TOKEN: "${GITHUB_TOKEN}"

RAG Pipeline Example

Ingest documents into PGVector, search by cosine similarity, feed context into AI.

name: rag-pipeline

tasks:
  - name: fetch-docs
    type: http.request
    properties:
      url: https://api.example.com/documents
      method: GET

  - name: embed-and-store
    type: ai.openai.embedding
    depends_on: [fetch-docs]
    properties:
      model: text-embedding-3-small
      input: '{{ outputs["fetch-docs"]["body"] }}'
      store: true
      collection: knowledge_base

  - name: search
    type: ai.openai.embedding
    depends_on: [embed-and-store]
    properties:
      model: text-embedding-3-small
      input: "What are the deployment requirements?"
      search: true
      collection: knowledge_base
      top_k: 5

  - name: answer
    type: ai.openai.chat
    depends_on: [search]
    properties:
      model: gpt-4o
      messages:
        - role: system
          content: "Answer based on this context only."
        - role: user
          content: |
            Context: {{ outputs["search"]["results"] }}
            Question: What are the deployment requirements?

Per-Task API Keys

Each task can use its own API key, or fall back to the environment variable:

# Uses OPENAI_API_KEY from environment
- name: default-key
  type: ai.openai.chat
  properties:
    model: gpt-4o-mini
    messages: [...]

# Uses a specific key (e.g. from credentials store)
- name: custom-key
  type: ai.openai.chat
  properties:
    model: gpt-4o
    api_key: "${PROJECT_OPENAI_KEY}"
    messages: [...]