Skip to content

README.md

Latest commit

 

History

History
171 lines (132 loc) · 6.88 KB

README.md

File metadata and controls

171 lines (132 loc) · 6.88 KB

Syrin examples

This directory contains example scripts showing how to use the Syrin library.

Get started quickly — one notebook

getting_started.ipynb — A single Jupyter notebook with all topics: markdown docs, runnable code, and output. Uses Almock (no API key). Sections: Minimal → Tasks → Budget (with thresholds) → Memory → Tools → Loops → Multi-agent → Streaming → Guardrails → Observability (trace/debug) → Context → Checkpoints → Models → Prompts → Advanced, plus a real-world section (budget + threshold + debug=True). Share this file with anyone who wants to get started quickly.

Example Categories

01_minimal/

Basic agent creation and simple responses.

  • hello_model.py — Create a Model and inspect config (no API key).
  • hello_agent.py — Agent with budget, prompt template (@prompt), and Response. Requires API key.
  • hello_memory.py — Agent with BufferMemory; multi-turn conversation with context.

02_tasks/

Task-based execution patterns.

  • single_task.py — Basic task execution.
  • multiple_tasks.py — Running multiple tasks.
  • task_with_output_type.py — Tasks with structured output.

03_budget/

Budget management and cost tracking.

  • Examples showing budget limits, thresholds, and cost tracking.

04_memory/

Memory systems and conversation history.

  • basic_memory.py — Buffer memory for multi-turn conversations.
  • async_memory.py — Async memory operations.
  • memory_types_and_decay.py — Different memory types with decay strategies.
  • chroma_memory.py, qdrant_memory.py, postgres_memory.py — Vector store backends.

05_tools/

Tool definitions and execution.

  • Tool creation, registration, and execution patterns.

06_loops/

Agent loop strategies (REACT, HITL, etc.).

  • react_loop.py — REACT pattern execution.
  • human_in_the_loop.py — Human approval workflows.

07_multi_agent/

Multi-agent patterns: handoff, spawn, pipeline, dynamic pipeline.

  • pipeline.py — Sequential/parallel pipeline execution.
  • dynamic_pipeline_basic.py — LLM-orchestrated multi-agent systems.
  • dynamic_pipeline_5agents.py — Complex multi-agent orchestration.

08_streaming/

Streaming responses and async patterns.

  • stream_sync.py, stream_async.py — Token-by-token output.

09_guardrails/

Content filtering, validation, and safety.

  • See 09_guardrails/README.md for details.

10_observability/

Tracing, debugging, and metrics.

  • comprehensive_tracing.py — Full observability setup.
  • audit_logging.py — Audit log recording.

11_context/

Context management, token limits, compaction.

  • context_management.py — Context configuration and thresholds.
  • context_snapshot_demo.py — Inspect context contents.
  • context_thresholds_compaction_demo.py — Automatic compaction.
  • context_runtime_injection_demo.py — RAG injection patterns.

11_mcp/

Model Context Protocol integration.

  • mcp_client.py — MCP client usage.

12_checkpoints/

Agent state persistence and recovery.

  • long_running_agent.py — Checkpointing for long sessions.

12_remote_config/

Remote configuration management.

  • init_and_serve.py — Remote config server.

13_models/

Model configuration and providers.

  • Model selection, configuration, and pricing.

14_prompts/

Prompt engineering and templates.

  • System prompts, user prompts, and validation.

15_advanced/

Advanced patterns: dependency injection, circuit breaker, agent inheritance.

  • dependency_injection.py — Passing dependencies to tools.
  • circuit_breaker.py — Circuit breaker pattern for LLM failures.
  • agent_inheritance.py — Extending Agent classes.
  • config.py — Configuration management.
  • hitl_approval.py — Human-in-the-loop approval.

16_serving/

HTTP serving and playground UI.

  • playground_dynamic_pipeline.py — Serve multi-agent pipelines.
  • See 16_serving/README.md for details.

17_routing/

Model routing and cost optimization.

  • model_router.py — Route to different models based on cost/quality.
  • See 17_routing/README.md for details.

18_multimodal/

Image, video, and voice generation.

  • image_generation.py — Generate images with DALL-E.
  • voice_generation.py — Generate speech with TTS providers.
  • See 18_multimodal/README.md for details.

19_knowledge/

Knowledge management and RAG (Retrieval-Augmented Generation).

  • loaders_and_document.py — Document loading from files, URLs, GitHub.
  • chunking.py — Text chunking strategies.
  • vector_store.py — Vector storage backends.
  • agentic_rag.py — Agentic RAG patterns.
  • full_rag_lifecycle.py — Complete RAG pipeline.
  • knowledge_agent.py — Agent with knowledge base.
  • postgres_backend.py — PostgreSQL vector store.
  • serve_agentic_postgres.py — Serve RAG agent.

20_template/

Template engine for slot-based generation.

  • template_standalone.py — Template with slots, render(), slot_schema(), from_file/from_string
  • template_with_agent.py — Agent with output_config.template; response.content = rendered, response.template_data = slot values
  • output_file_generation.py — output_config format (TEXT/MARKDOWN/PDF/DOCX); citation parsing (CitationConfig); response.file, response.file_bytes, response.citations; save_as_pdf, save_as_docx

ipo_drafting_agent/

Real-world example: DRHP Drafting Agent for Capital Structure & Shareholding Pattern section.

  • run.py — Single command to draft a DRHP section from ROC filings (PAS-3, SH-7, MOA, etc.)
  • Uses Knowledge (RAG), agentic retrieval, structured output, automation transparency
  • Outputs: intermediate JSON, draft Markdown/DOCX, sources used, items requiring review

resume_agent/

Real-world example: Voice AI agent for recruiter calls using Syrin + Pipecat.

Running examples

From the project root (with OPENAI_API_KEY or other provider keys in examples/.env):

# Run a specific example
PYTHONPATH=. python examples/01_minimal/hello_agent.py

# Run a module
PYTHONPATH=. python -m examples.01_minimal.hello_agent

Setup

  1. From project root with virtualenv activated:

    source .venv/bin/activate
uv pip install -e ".[dev,anthropic]"

  2. Add a .env file in this directory (examples/.env) with your API key:

    ANTHROPIC_API_KEY=sk-ant-...

    Optional: ANTHROPIC_MODEL_ID=anthropic/claude-3-7-sonnet-latest (default; use any valid id from Anthropic's API).

Examples load .env via python-dotenv. Run: python examples/01_minimal/hello_agent.py

CLI (transport): Syrin serve -a examples.hello_agent:agent (interactive REPL), Syrin cost, Syrin version, Syrin run <script.py>.

Remote config

  • 12_remote_config/ — syrin.init(), config routes (GET/PATCH /config, /config/stream) when serving. See 12_remote_config/README.md.