Skip to main content

​
Install

pip install synap-microsoft-agent

​
What’s included

ClassPurpose
SynapContextProviderInjects Synap memory into the agent’s context before each turn, and records turns after
SynapHistoryProviderPersists and loads full conversation history via Synap

​
Quick start

from azure.ai.agents import AgentsClient
from synap_microsoft_agent import SynapContextProvider, SynapHistoryProvider

client = AgentsClient(endpoint=os.environ["AZURE_AI_ENDPOINT"])

agent = client.as_agent(
    context_providers=[
        SynapContextProvider(
            sdk=sdk,
            user_id="alice",
            customer_id="acme",   # optional
            max_context_results=6,
        ),
        SynapHistoryProvider(
            sdk=sdk,
            user_id="alice",
            conversation_id="thread-001",
        ),
    ],
)

response = await agent.run("What were the outcomes from my last meeting?")

​
SynapContextProvider

Called by the MAF runtime at the start of each turn:
  1. Fetches relevant memories for the incoming message
  2. Appends them as a system context message
  3. After the agent responds, ingests the full turn back into Synap
Failures on step 1 degrade gracefully (empty context, error logged). Failures on step 3 raise so callers know persistence failed.

​
SynapHistoryProvider

Persists and reloads the full conversation message list:
  • load(thread_id) — fetches prior messages from Synap and restores them to the MAF thread
  • save(thread_id, messages) — writes new messages to Synap after each turn
Use SynapHistoryProvider when you need the LLM to see the full conversation transcript, not just semantic context snippets.

​
Using both together

agent = client.as_agent(
    context_providers=[
        SynapContextProvider(sdk=sdk, user_id="alice", customer_id="acme"),
        SynapHistoryProvider(sdk=sdk, user_id="alice", conversation_id="thread-001"),
    ],
)
SynapContextProvider handles semantic memory injection; SynapHistoryProvider handles verbatim history recall. Use both for agents that need both types of memory.

​
Next steps

Semantic Kernel

Kernel plugin for Microsoft Semantic Kernel.

NeMo Agent Toolkit

MemoryEditor for NVIDIA NeMo workflows.