Configure a provider

Goal: wire Noeta to a real LLM endpoint — Anthropic, OpenAI-compatible, or OpenAI Responses API.

Before you start: you have installed Noeta and run the stub provider from the quickstart. You have an API key for your chosen provider.

Option A: environment variables (coding agent)

The easiest way to configure a provider for python -m noeta.agent is through NOETA_AGENT_* environment variables:

# Anthropic
NOETA_AGENT_PROVIDER=anthropic \
NOETA_AGENT_MODEL=claude-sonnet-4-5-20250929 \
NOETA_AGENT_API_KEY=sk-ant-… \
python -m noeta.agent

# OpenAI-compatible (any endpoint that speaks the OpenAI chat/completions format)
NOETA_AGENT_PROVIDER=openai \
NOETA_AGENT_MODEL=gpt-5.5 \
NOETA_AGENT_BASE_URL=https://api.openai.com/v1 \
NOETA_AGENT_API_KEY=sk-… \
python -m noeta.agent

# OpenAI Responses API
NOETA_AGENT_PROVIDER=openai-responses \
NOETA_AGENT_MODEL=gpt-5.5 \
NOETA_AGENT_API_KEY=sk-… \
python -m noeta.agent

NOETA_AGENT_BASE_URL is optional for the official OpenAI endpoint (defaults to https://api.openai.com/v1); set it explicitly for self-hosted or third-party OpenAI-compatible endpoints.

Option B: JSON config file

Instead of individual env vars, you can point to a JSON config file:

NOETA_AGENT_CONFIG=./noeta-config.json python -m noeta.agent

Where noeta-config.json contains:

{
  "provider": "anthropic",
  "model": "claude-sonnet-4-5-20250929",
  "api_key": "sk-ant-…",
  "workspace": "./my-project",
  "storage_url": "./session.sqlite",
  "write_mode": "dry_run"
}

Option C: programmatic (SDK)

When using the SDK directly (noeta.sdk), pass a provider to Options or Client:

from noeta.sdk import Client, Options
from noeta.llm.anthropic import AnthropicProvider

provider = AnthropicProvider(
    model="claude-sonnet-4-5-20250929",
    api_key="sk-ant-…",
)

options = Options(
    system_prompt="You are a helpful assistant.",
    name="my-agent",
    provider=provider,
)

client = Client(options, workspace_dir="./workspace")

For OpenAI-compatible endpoints, use OpenAICompatProvider:

from noeta.llm.openai_compat import OpenAICompatProvider

provider = OpenAICompatProvider(
    model="gpt-5.5",
    base_url="https://api.openai.com/v1",
    api_key="sk-…",
)

And for the OpenAI Responses API:

from noeta.llm.openai_responses import OpenAIResponsesProvider

provider = OpenAIResponsesProvider(
    model="gpt-5.5",
    api_key="sk-…",
)

Verify it works

Boot the agent with your real provider and send a simple message:

NOETA_AGENT_PROVIDER=anthropic \
NOETA_AGENT_MODEL=claude-sonnet-4-5-20250929 \
NOETA_AGENT_API_KEY=sk-ant-… \
NOETA_AGENT_STORAGE=./test.sqlite \
python -m noeta.agent

Open the chat UI and ask "What is 2 + 2?". If you get a real LLM response (rather than the stub's canned reply), the provider is wired correctly. Check the trace view to confirm the turn shows real token counts and usage.

Switching providers

To swap backends, change the provider instance — no other code needs to change. See Swap providers for a before/after example. The internal protocol is vendor-neutral, so your agent code, tools, and recorded history are portable across providers.

All three built-in providers stream tokens to the web UI while a response is being generated — no configuration needed. The stub provider does not stream (it answers instantly), so don't expect a typing effect on a bare boot.

Troubleshooting

• "No provider configured" — NOETA_AGENT_PROVIDER is unset (defaults to stub) or Options.provider is None.
• 401 / authentication error — check your API key. For Anthropic, it starts with sk-ant-; for OpenAI, sk-.
• Model not found — verify the model name. Anthropic model names include the date suffix (e.g. claude-sonnet-4-5-20250929).
• Connection timeout — check NOETA_AGENT_BASE_URL. For corporate proxies, you may need to set HTTPS_PROXY.

See Troubleshooting for more.

See also

• Provider neutrality — why the internal protocol is vendor-agnostic
• Swap providers — before/after code example
• SDK reference — Options and Client constructor parameters