Core Feature

Multi-Provider Orchestration

One workflow definition. Any LLM provider. Zero lock-in.

The LLM Market in 2025

Enterprise LLM adoption has shifted dramatically. Based on Menlo Ventures research:

40%

Anthropic

claude-opus-4, claude-sonnet-4-5

27%

OpenAI

gpt-4o, gpt-4-turbo

21%

Google

gemini-2.0-flash, gemini-1.5-pro

Growing

Local (Ollama)

llama3.2, mistral

Key insight: No single provider dominates. Enterprise teams use 2-3 providers on average. Your workflow tool should support this reality.

How It Works in Nika 

# Define providers once at the top
providers:
  anthropic:
    type: anthropic
    apiKeyEnv: ANTHROPIC_API_KEY
  openai:
    type: openai
    apiKeyEnv: OPENAI_API_KEY
  local:
    type: ollama
    baseUrl: http://localhost:11434

# Use different models per task
tasks:
  - id: analyze
    agent:
      prompt: "Deep code analysis"
      model: claude-opus-4        # Best for reasoning
      provider: anthropic

  - id: generate-docs
    agent:
      prompt: "Write documentation"
      model: gpt-4o              # Good for writing
      provider: openai

  - id: quick-check
    agent:
      prompt: "Quick syntax check"
      model: llama3.2            # Free, fast, local
      provider: local

Each task can use a different provider and model. The runner handles authentication, API differences, and response normalization transparently.

Why This Matters

No Vendor Lock-in

Switch providers without rewriting workflows. Same YAML, different model.

Cost Optimization

Use cheaper models for simple tasks, powerful ones for complex reasoning.

Resilience

If one provider is down, failover to another automatically.

Best Tool for the Job

Claude for coding, GPT for creative, Gemini for multimodal, local for privacy.

Future: Intelligent Routing (Research)

Based on the "Pick and Spin" research framework (arXiv, Dec 2024), we're exploring automatic model routing that optimizes for: 

# Potential future syntax (not implemented yet)
agent:
  prompt: "Complex analysis task"
  routing:
    strategy: auto           # Let Nika choose
    optimize:
      - relevance            # Best model for this task
      - latency              # Fastest response
      - cost                 # Cheapest option
    fallback: [claude-sonnet-4-5, gpt-4o]

# Research shows this can achieve:
# - 21.6% higher success rates
# - 30% lower latency
# - 33% lower cost per query

Status: This is research territory. We're evaluating whether automatic routing adds enough value to justify the complexity. Your feedback matters.

Compatibility: Model Context Protocol (MCP)

Anthropic's MCP (announced Nov 2024) is becoming an industry standard for LLM tool integration. Nika supports MCP for tool invocation: 

# Invoke MCP-compatible tools
tasks:
  - id: search-docs
    invoke:
      reference: "mcp::docs-search::query"
      args:
        query: "authentication flow"

# Works with any MCP server:
# - Anthropic's reference servers
# - Third-party integrations
# - Your custom MCP servers

Multi-provider support is live

Use any model in your workflows.

All Concepts Join Waitlist