The context-based multi‑agent framework. Build agents that share, route, and persist context — with first‑class tooling for prompts, tools, and popular LLM frameworks.
- Context Mesh with topic routing and user isolation
- ToolHandler with adapters (OpenAI, Anthropic, LangChain, Agno)
- Prompt builders for system and message prompts
- Pluggable persistence (SQLite, PostgreSQL)
- Lightweight, framework‑agnostic, production‑ready
pip install synthafrom syntha import ContextMesh, ToolHandler, build_system_prompt
# 1) Shared context, isolated per user
context = ContextMesh(user_id="demo_user")
# 2) Agents interact via tools (no manual data passing)
handler = ToolHandler(context, "AssistantAgent")
context.push("project", "AI Customer Support")
context.push("status", "active", topics=["support"]) # topic‑routed
# 3) Context‑aware prompts for your LLM
system_prompt = build_system_prompt("AssistantAgent", context)
print(system_prompt[:200] + "...")- OpenAI function calling:
handler.get_openai_functions() - Anthropic tool use:
handler.get_anthropic_tools() - LangChain BaseTool:
handler.get_langchain_tools() - Agno Functions:
handler.get_tools_for_framework("agno")
See the docs for concise, copy‑paste examples.
from syntha import ContextMesh, ToolHandler
mesh = ContextMesh(user_id="demo")
handler = ToolHandler(mesh, agent_name="Assistant")
agno_tools = handler.get_tools_for_framework("agno")
try:
from agno.agent import Agent
agent = Agent(
name="Assistant",
tools=agno_tools,
instructions="Use tools to read/write context.",
model="gpt-4o",
)
# response = agent.run("List context keys and fetch 'project'")
except ImportError:
print("pip install agno to enable Agent integration")- Docs: https://doc.syntha.ca
- Quick Start: https://doc.syntha.ca/user-guide/introduction/quick-start/
- Examples: https://doc.syntha.ca/examples/overview/
- API Reference: https://doc.syntha.ca/api/overview/
Apache 2.0 © Syntha. See LICENSE.