ai-memory-mcp 0.6.4

Python SDK for the ai-memory HTTP API (persistent memory for AI agents).

ai-memory — Python SDK

Typed Python client for the ai-memory HTTP API. Wraps the daemon's /api/v1/ surface with sync and async clients, Pydantic v2 models that mirror the Rust structs, and HMAC-SHA256 webhook verification.

Status: 0.6.0-alpha.0 — scaffolding, unpublished. API may change before GA.

Install

# Not yet published; install from a local checkout:
pip install -e ./sdk/python

Requires Python 3.10+.

Quickstart

from ai_memory import AiMemoryClient, Tier

with AiMemoryClient(base_url="http://localhost:9077") as client:
    created = client.store(
        title="BIND9 build notes",
        content="Use --with-openssl=/opt/openssl, disable DoH for the lab.",
        tier=Tier.LONG,
        tags=["dns", "bind9"],
    )
    print(created["id"])

    hits = client.recall(context="how do I build BIND9?")
    for memory in hits.memories:
        print(memory.title, memory.confidence)

Async

import asyncio
from ai_memory import AsyncAiMemoryClient

async def main() -> None:
    async with AsyncAiMemoryClient(base_url="http://localhost:9077") as client:
        resp = await client.recall(context="hello")
        for memory in resp.memories:
            print(memory.title)

asyncio.run(main())

Authentication

API key

AiMemoryClient(base_url="https://memory.example.com", api_key="sk-...")

The key is sent as the X-API-Key header on every request. The server exempts /api/v1/health from auth.

mTLS

AiMemoryClient(
    base_url="https://memory.example.com",
    verify="/etc/ssl/certs/server-ca.pem",
    cert=("/etc/ssl/client/client.pem", "/etc/ssl/client/client.key"),
)

Agent identity (NHI)

Set agent_id to stamp the X-Agent-Id header on every request. The server writes metadata.agent_id accordingly (see CLAUDE.md §Agent Identity).

AiMemoryClient(base_url="http://localhost:9077", agent_id="ai:claude-opus-4.7@host")

All methods

Method	Endpoint	Notes
health()	GET /api/v1/health	Exempt from auth.
metrics()	GET /api/v1/metrics	Prometheus text format.
store(...)	POST /api/v1/memories	Upsert on (title, namespace).
bulk_store([...])	POST /api/v1/memories/bulk	Up to 1000 per call.
get(id)	GET /api/v1/memories/{id}	Returns Memory.
update(id, UpdateMemory)	PUT /api/v1/memories/{id}	Partial update.
delete(id)	DELETE /api/v1/memories/{id}
promote(id)	POST /api/v1/memories/{id}/promote	Tier -> long.
list(...)	GET /api/v1/memories	Filters: namespace, tier, tags, agent_id.
search(q, ...)	GET /api/v1/search	FTS AND search.
recall(context, ...)	POST /api/v1/recall	Hybrid semantic + FTS.
forget(...)	POST /api/v1/forget	Bulk delete by pattern.
link(a, b, relation)	POST /api/v1/links	related_to, supersedes, contradicts, derived_from.
get_links(id)	GET /api/v1/links/{id}
stats()	GET /api/v1/stats
namespaces()	GET /api/v1/namespaces
gc()	POST /api/v1/gc
export() / import_()	GET / `POST /api/v1/export	import`
subscribe(req) / unsubscribe(id) / subscriptions()	/api/v1/subscriptions	Webhook mgmt.
notify(req) / inbox(...)	/api/v1/notify, /api/v1/inbox	Agent-to-agent messaging.
grant(id, agent) / revoke(id, agent)	`/api/v1/memories/{id}/grant	revoke`
cluster(req)	POST /api/v1/cluster	Peer management.
agents() / register_agent(...)	/api/v1/agents	NHI registry.

Webhook verification

from ai_memory import verify_webhook_signature

def handle(request) -> None:
    sig = request.headers["X-AI-Memory-Signature"]
    if not verify_webhook_signature(request.body, sig, secret="..."):
        raise PermissionError("bad signature")
    ...

body must be the raw bytes as received — do not re-encode a parsed JSON payload; whitespace differences will break the HMAC.

Errors

All SDK errors derive from AiMemoryError:

Exception	HTTP
ValidationError	400
AuthError	401
ForbiddenError	403
NotFoundError	404
ConflictError	409
RateLimitError	429
ServerError	5xx
TransportError	network failure

License

Apache-2.0, see LICENSE.