Skip to content

README.md

Latest commit

 

History

History
74 lines (53 loc) · 3.41 KB

README.md

File metadata and controls

74 lines (53 loc) · 3.41 KB
Raw

Python examples

The fastest way to understand Agent RT is to run it. These small examples are deliberately incremental: start with one agent call, then add tools, streaming, structured output, runtime limits, and direct registry control without introducing a second framework or hidden abstraction.

If you are evaluating Agent RT, run 01_basic_agent.py first, then 02_tool_calling.py. Together they show the core model → tool → model loop with the same contracts used by the larger runtime.

Examples

  • 01_basic_agent.py — minimal agent run with a configured provider.
  • 02_tool_calling.py — register a read-only tool and let the model call it.
  • 03_streaming.py — consume text deltas with run_streaming().
  • 04_structured_output.py — require schema-validated JSON output with one repair attempt.
  • 05_run_limits.py — bound turns, tool calls, elapsed time, and total tokens.
  • 06_tool_registry.py — use namespaces, direct execution, and enable/disable controls without a model provider.
  • 07_vector_db.py — query Chroma, Milvus, Pinecone, Qdrant, or Weaviate and switch backends with environment variables.

Setup

From the python/ directory:

uv sync

Or with pip:

python -m venv .venv
source .venv/bin/activate
python -m pip install -e .

Configure a provider. For OpenAI or an OpenAI-compatible endpoint:

export OPENAI_MODEL="your-model"
export OPENAI_API_KEY="..."
# Optional:
export OPENAI_BASE_URL="https://your-provider.example/v1"

For Anthropic, configure ANTHROPIC_BASE_URL, ANTHROPIC_API_KEY, and ANTHROPIC_MODEL.

Run an example from the python/ directory:

uv run python examples/01_basic_agent.py
uv run python examples/02_tool_calling.py
uv run python examples/03_streaming.py
uv run python examples/04_structured_output.py
uv run python examples/05_run_limits.py
uv run python examples/06_tool_registry.py
uv run python examples/07_vector_db.py

The first five examples call a configured model provider. 06_tool_registry.py is fully offline and is useful for checking local setup without API credentials. 07_vector_db.py performs live vector retrieval and requires a configured vector service plus an embedding-capable model provider.

Vector DB example

The vector DB example uses the same application code for every built-in service backend. Configure an OpenAI/OpenAI-compatible embedding provider, then set:

export AGENT_RT_VECTOR_DB="qdrant" # chroma|milvus|pinecone|qdrant|weaviate
export AGENT_RT_VECTOR_DB_URL="https://your-vector-db.example"
export AGENT_RT_VECTOR_DB_COLLECTION="docs"
export AGENT_RT_VECTOR_DB_API_KEY="..." # optional; backend-native variables also work
export AGENT_RT_VECTOR_DB_QUERY="What is Agent RT?" # optional
export AGENT_RT_VECTOR_DB_LIMIT="5" # optional

Backend-native credentials are CHROMA_API_KEY, MILVUS_TOKEN, PINECONE_API_KEY, QDRANT_API_KEY, and WEAVIATE_API_KEY. Change AGENT_RT_VECTOR_DB plus the endpoint, collection, and credential values to move the same application code to another backend. Extra backend configuration is passed with AGENT_RT_VECTOR_DB_OPTION_* variables, for example AGENT_RT_VECTOR_DB_OPTION_NAMESPACE for Pinecone or AGENT_RT_VECTOR_DB_OPTION_TENANT / AGENT_RT_VECTOR_DB_OPTION_DATABASE for Chroma.

FAISS and pgvector remain custom VectorDBProviderRegistry integrations because their normal execution paths require local/native or database drivers.