Local-First Run

Zero-config first run that picks a model from the provider credentials you already have

The terminal CLI picks a sensible default model from whichever provider credential is present, so praisonai run "..." works without passing --model.

from praisonaiagents import Agent

agent = Agent(name="local-first", instructions="Answer clearly and briefly.")
agent.start("Explain black holes to a 5-year-old")

Quick Start

1. Point at a local model (optional)

   Run an OpenAI-compatible endpoint such as Ollama, then export its host.

   ollama pull llama3.2 && ollama serve
   export OLLAMA_HOST=http://127.0.0.1:11434

2. Run with no –model

   The CLI resolves a provider-appropriate default from your environment.

   praisonai run "Explain black holes to a 5-year-old"

3. Override the endpoint (optional)

   Point at any OpenAI-compatible server without editing config.

   export OPENAI_BASE_URL=http://localhost:11434/v1
   praisonai run "Summarise today's news"

4. Same path from the SDK

   The SDK Agent() constructor resolves the same default — no llm=, only OLLAMA_HOST set. This works from the SDK as of PR #4795.

   # no OPENAI_API_KEY, no llm=; only OLLAMA_HOST is set
   from praisonaiagents import Agent
   
   Agent(instructions="local-first").start("Explain black holes to a 5-year-old")

How It Works

The CLI checks for a credential, then resolves a default model before running.

is_configured() treats a run as ready when any provider key is set (env var or stored credential). resolve_default_model() then chooses the model, persisting only user-chosen models for next time.

Detection Order

The default model follows a fixed precedence.

Provider-inferred default models per credential:

When a provider-inferred default is used, the CLI prints one transparency line, for example: No model set; using ollama/llama3.2 because OLLAMA_HOST is present.

Inspect which rung won

Not sure which model your next run will use? Ask the CLI:

praisonai config provenance run.model
# run.model = ollama/llama3.2
#   layer: model-resolver  source: local

The source tag maps directly to the detection-order rungs above (state:, env:MODEL_NAME, local, provider:…, fallback). See praisonai config provenance for the full tag list.

Credential Gate

The run behaves differently when no credential is found, depending on whether the terminal is interactive.

Endpoint Environment Variables

Point the CLI at any OpenAI-compatible endpoint without editing config files.

Best Practices

Related