OpenClaw Integration

Use Papr as the memory layer for OpenClaw agents so chat, documents, and structured events remain connected across runs.

What You Build

A Papr-backed memory adapter for OpenClaw workflows
User-scoped memory retrieval before each task
Post-task writeback for new facts and agent learnings

Prerequisites

Papr API key in PAPR_MEMORY_API_KEY
OpenClaw runtime with configurable memory adapter
Stable external_user_id from your application identity

Integration Pattern

Pre-task retrieval: query Papr with the current goal.
Execution: run OpenClaw task using retrieved context.
Post-task writeback: store outcomes, decisions, and extracted entities.
Optional analytics: run GraphQL queries for relationship-level insights.

Minimal Setup

Configure a Papr adapter in your OpenClaw runtime.
Call search before each task to preload memory context.
Call add after each task to persist outcomes.
Validate retrieval quality with a repeated test scenario.

Python Adapter Skeleton

import os
from dataclasses import dataclass
from papr_memory import Papr


@dataclass
class OpenClawContext:
    external_user_id: str
    organization_id: str
    namespace_id: str


class PaprOpenClawMemoryAdapter:
    def __init__(self) -> None:
        self.client = Papr(x_api_key=os.environ.get("PAPR_MEMORY_API_KEY"))

    def pre_task_context(self, goal: str, ctx: OpenClawContext):
        return self.client.memory.search(
            query=goal,
            external_user_id=ctx.external_user_id,
            organization_id=ctx.organization_id,
            namespace_id=ctx.namespace_id,
            enable_agentic_graph=True,
            max_memories=20,
            max_nodes=15,
        )

    def post_task_learning(self, summary: str, ctx: OpenClawContext):
        return self.client.memory.add(
            content=summary,
            external_user_id=ctx.external_user_id,
            organization_id=ctx.organization_id,
            namespace_id=ctx.namespace_id,
            metadata={"role": "assistant", "category": "learning"},
            memory_policy={"mode": "auto"},
        )

Production Settings

Use one namespace per environment (dev, staging, prod)
Keep write payloads short and factual to improve retrieval precision
Store task IDs in metadata for reproducible debugging
Keep enable_agentic_graph=true for multi-hop retrieval paths

Validation Checklist

Retrieval returns relevant context before task execution
Post-task writes produce new searchable memory within expected latency
Cross-run sessions preserve agent learning for the same user
Tenant boundaries are respected across namespaces

Troubleshooting

If retrieval is weak or writes do not appear, confirm payload scope fields and use Error Playbook.