Works with Every
AI Framework
AISP is a transport and session layer — not a framework. It works alongside LangChain, AutoGen, CrewAI, and any other framework. Add persistent sessions and shared memory in under 10 lines.
3-Step Integration Guide
pip install aisp or go get github.com/blackholemesh/aisp-go
pip install aispEach agent gets a unique session. Sessions survive restarts.
session = AISPSession(agent_id="my-agent")Read and write across agents in the same realm.
session.memory.set("key", value)Any framework, any language
Wrap any LangChain agent with an AISP session. Memory namespaces replace chain memory — persistent across runs.
from aisp import AISPSession
session = AISPSession(agent_id="lc-agent-1")
session.memory.set("context.user", "Jarji")
# Use with any LangChain chain
chain.run(memory=session.memory)Give AutoGen agents shared memory and persistent sessions. Agents in a GroupChat can read each other's namespaces.
from aisp import AISPMemory
memory = AISPMemory(namespace="task-autogen-1")
memory.set("plan", "Analyze quarterly report")
agent = AssistantAgent(
name="analyst",
aisp_memory=memory
)AISP sessions map naturally to CrewAI crews. Each crew member gets a session; shared crew memory lives in the default namespace.
from aisp import AISPSession
session = AISPSession(agent_id="crew-researcher")
session.memory.set("crew.goal", "Market analysis")
@agent.task
def research(ctx):
goal = session.memory.get("crew.goal")
return analyze(goal)Replace OpenAI thread storage with AISP memory. Sessions persist beyond API rate limits and token windows.
from aisp import AISPSession
session = AISPSession(agent_id="oai-assistant-1")
# Store assistant thread state
session.memory.set("thread_id", thread.id)
session.memory.set("messages", messages)
# Retrieve on next run — even after restart
thread_id = session.memory.get("thread_id")AISP extends Claude with persistent session identity. Multiple Claude agents coordinate via shared CRDT memory.
from aisp import AISPSession
import anthropic
session = AISPSession(agent_id="claude-agent-1")
client = anthropic.Anthropic()
ctx = session.memory.get("context") or {}
response = client.messages.create(
model="claude-opus-4-5",
system=f"Context: {ctx}",
messages=[{"role": "user", "content": prompt}]
)The Go SDK is the reference implementation. Full protocol support including ZK proofs, CRDT memory, and peer discovery.
import "github.com/blackholemesh/aisp-go"
sess, err := aisp.NewSession(ctx, agentID)
if err != nil { log.Fatal(err) }
defer sess.Close()
// Set memory
sess.Memory().Set("task.result", result)
// Send message to peer
sess.Send(peerID, aisp.Message{Content: "done"})Ready to integrate?
Install the SDK, create a session, and your first agent is running in under 5 minutes.