forgememory-enterprise 1.0.0

Python SDK for ForgeMemory Enterprise — multi-tenant AI memory at scale

ForgeMemory Enterprise Python SDK

Production-quality Python client for ForgeMemory Enterprise — multi-tenant AI memory at scale.

Installation

pip install forgememory-enterprise

With framework adapters:

pip install "forgememory-enterprise[crewai]"     # CrewAI integration
pip install "forgememory-enterprise[langgraph]"   # LangGraph integration
pip install "forgememory-enterprise[autogen]"     # AutoGen integration

Quick Start

from forgememory_enterprise import EnterpriseClient
from forgememory_enterprise.types import CreateMemoryRequest

client = EnterpriseClient(api_key="fme_your_key", base_url="https://your-server:8201")

# Store a memory
mem = client.memory.create("tenant-123", CreateMemoryRequest(
    content="Q4 revenue target is $2.5M",
    memory_type="semantic",
    importance=8,
    tags=["finance"],
))

# Search memories
results = client.memory.search("tenant-123", "revenue target", limit=5)

# Get assembled LLM context
ctx = client.memory.context("tenant-123", "What are our targets?")
print(ctx.context)

Server Address

base_url is a required parameter — the SDK never embeds server addresses:

client = EnterpriseClient(api_key="fme_your_key", base_url="https://fme.example.com:8201")

For deployments that need dynamic LAN/Tailscale routing, use _network.detect_base_url() and pass the result as base_url. The detection function requires caller-supplied host names — no defaults are baked in:

from forgememory_enterprise._network import detect_base_url

url = detect_base_url(lan_host="your-lan-host", tailscale_host="your-ts-host")
client = EnterpriseClient(api_key="fme_your_key", base_url=url)

For the MCP server, set one of:

• FME_BASE_URL=https://your-server:8201 (preferred)
• FME_LAN_HOST=<ip> + FME_TAILSCALE_HOST=<hostname> (auto-detection)

Sub-Managers

Manager	Access	Operations
client.orgs	OrgManager	Create/list orgs, users, branches, provision tenants
client.tenants	TenantManager	List, suspend, reactivate, delete, set quota
client.api_keys	ApiKeyManager	Create, list, revoke API keys
client.agents	AgentRegistry	Register, update, delete agents; sessions, heartbeat
client.audit	AuditLogger	Log actions, memory access, delegations; costs
client.memory	MemoryClient	CRUD, search, context, facts (per-tenant)
client.admin	AdminClient	Health, dashboard, pool stats
client.entities	EntityGraphClient	Entity CRUD, edges, BFS graph traversal
client.ingest	IngestClient	Document, email, meeting, calendar ingestion

All managers are lazily initialized on first access.

Agent Sessions

with client.session("agent-1", "tenant-1") as sess:
    # Background heartbeat runs automatically
    client.audit.log_action("agent-1", sess.id, "tool_call", "web_search")
    client.memory.create("tenant-1", CreateMemoryRequest(content="Found info"))
# Session auto-ended (status=completed, or failed on exception)

Framework Adapters

CrewAI

from forgememory_enterprise.adapters.crewai_adapter import ForgeMemoryTool, FMECrewObserver

tool = ForgeMemoryTool(client, tenant_id="t-123")
result = tool.run("search:customer preferences")

observer = FMECrewObserver(client, "agent-1", session_id)
observer.on_tool_use("search", "query")
observer.on_llm_call(tokens_in=500, tokens_out=200, cost_usd=0.005)

LangGraph

from forgememory_enterprise.adapters.langgraph_adapter import (
    make_memory_retrieve_node, make_memory_store_node, make_memory_context_node
)

retrieve = make_memory_retrieve_node(client, "t-123")
store = make_memory_store_node(client, "t-123")
# Add to your graph as nodes

AutoGen

from forgememory_enterprise.adapters.autogen_adapter import fme_tool, inject_context

@fme_tool(client, tenant_id="t-123")
def search_memory(query: str) -> str:
    """Search organizational memory."""
    ...

context = inject_context(client, "t-123", "project status")

OpenClaw

from forgememory_enterprise.adapters.openclaw_adapter import FMEOpenClawPlugin

plugin = FMEOpenClawPlugin(client, tenant_id="t-123")
results = plugin.search("customer preferences")
plugin.store("New finding from browsing session")

Error Handling

from forgememory_enterprise import (
    AuthenticationError,   # 401 — invalid/expired key
    ForbiddenError,        # 403 — insufficient permissions
    NotFoundError,         # 404 — resource not found
    BudgetExceededError,   # 402 — agent budget exceeded
    RateLimitedError,      # 429 — too many requests
    LegalHoldError,        # 403 — GDPR legal hold blocks mutation
)

Testing

cd sdk/python
pip install -e ".[dev]"
python -m pytest tests/ -v  # 367 tests, all using MockTransport

Requirements

• Python 3.10+
• httpx >= 0.27.0
• pydantic >= 2.0