piqrypt-langchain-integration 1.0.0

PiQrypt Verifiable AI Agent Memory bridge for LangChain

piqrypt-langchain

Verifiable AI Agent Memory for LangChain.

Every tool call, LLM response, chain execution, and agent action — signed, hash-chained, tamper-proof.

pip install piqrypt-langchain

---

Quickstart — attach once, stamp everything

from langchain_openai import ChatOpenAI
from langchain.agents import AgentExecutor, create_openai_tools_agent
from piqrypt_langchain import PiQryptCallbackHandler

# One handler — stamps every LLM call, tool call, and chain event
handler = PiQryptCallbackHandler(identity_file="my-agent.json")

llm = ChatOpenAI(model="gpt-4o", callbacks=[handler])
agent = create_openai_tools_agent(llm, tools, prompt)
executor = AgentExecutor(agent=agent, tools=tools, callbacks=[handler])

result = executor.invoke({"input": "Analyze Q4 sales and flag anomalies"})

# Export verifiable memory
handler.export_audit("q4-analysis-audit.json")
# $ piqrypt verify q4-analysis-audit.json

---

Drop-in AgentExecutor

from piqrypt_langchain import AuditedAgentExecutor

# Replace AgentExecutor with AuditedAgentExecutor
executor = AuditedAgentExecutor(
    agent=your_agent,
    tools=your_tools,
    identity_file="my-agent.json"
)

result = executor.invoke({"input": "Your query here"})
executor.export_audit("audit.json")

---

Wrap individual tools

from langchain.tools import tool
from piqrypt_langchain import piqrypt_tool

@tool
@piqrypt_tool(identity_file="my-agent.json")
def search_web(query: str) -> str:
    """Search the web for information."""
    return your_search_logic(query)

@tool
@piqrypt_tool(identity_file="my-agent.json")
def execute_sql(query: str) -> str:
    """Execute a SQL query."""
    return your_db_logic(query)

---

Wrap chain functions

from piqrypt_langchain import stamp_chain

@stamp_chain("document_analysis", identity_file="my-agent.json")
def analyze_document(doc: str) -> dict:
    return your_chain.invoke({"input": doc})

---

What gets stamped

Event	When
callback_handler_initialized	Handler creation
llm_response	After every LLM call
tool_start	Before tool execution
tool_end	After tool execution
tool_error	On tool failure
chain_start	Before chain runs
chain_end	After chain completes
agent_action	Every agent decision
agent_finish	Agent completion
executor_invoke	AgentExecutor input
executor_complete	AgentExecutor output

All events Ed25519-signed, SHA-256 hash-chained.
Raw inputs and outputs never stored — only their SHA-256 hashes.

---

Verify

piqrypt verify langchain-audit.json
# ✅ Chain integrity verified — 32 events, 0 forks

piqrypt search --type tool_error
# All tool failures with timestamps

---

Why the callback handler is the best approach

LangChain's callback system is designed exactly for this — attach once at the top level, every nested component fires the same callbacks. One PiQryptCallbackHandler stamps your entire agent pipeline automatically, including:

• Every LLM call inside chains
• Every tool invocation
• Every intermediate chain step
• Every agent decision

No need to modify individual tools or chains.

---

Scope

Use case	AISS profile
Development / PoC	AISS-1 (Free, included)
Non-critical production	AISS-1 (Free)
Regulated production	AISS-2 (Pro)

---

Links

• PiQrypt core: github.com/piqrypt/piqrypt
• Integration guide: INTEGRATION.md — LangChain
• Issues: [e-mail]

---

PiQrypt — Verifiable AI Agent Memory