OpenClaw-Native Memory & Context Runtime
Elephants
never forget.
Purpose-built for OpenClaw. Portable to any agent framework. ElephantBroker gives AI agents persistent memory, intelligent context management, and enterprise-grade safety — running entirely on your infrastructure if you choose.
The Problem
Your AI isn't smart until it remembers.
Every session, your agents start from scratch. Chat history gets cut off. Search retrieves fragments but can't tell what's important. Databases store data but lose the connections between ideas. And when multiple agents collaborate? They can't share what they've learned.
The result? Agents that repeat mistakes, forget your preferences, violate procedures you've already defined, and lose track of complex, multi-step work across teams.
Imagine this:
Session 1: "Never deploy to production on Fridays. That's our policy."
Session 47: "Deploying to production now — it's Friday at 4pm."
Without ElephantBroker, your agent forgot the rule. With it, a red-line guard blocks the action.
How It Works
Context management,
brokered intelligently.
Remember Everything
Every conversation, decision, and artifact is automatically saved and organized — categorized into facts, procedures, decisions, and evidence. Procedural memory gets its own dedicated repository, so your agent always knows how things should be done.
Your deployment runbook is stored as a procedure. When an agent tries to deploy, it follows the steps — every time, without being told.
A Living Context Window
Forget static context that gets compacted and thrown away. ElephantBroker maintains a living, sliding window that continuously adapts to the current goal — always surfacing what's relevant, always making room for what matters next.
Debugging auth? The context automatically fills with your security decisions, token configs, and the last three related bugs — not yesterday's CSS discussion.
Share or Isolate — You Decide
Multiple agents can share the same memory and context runtime. Knowledge can flow between agents or be isolated per session, per actor, per team — all configurable through policies. Built-in presets get you started in minutes.
A research agent discovers a new API endpoint. Your coding agent picks it up immediately — because they share the same knowledge plane.
Get Smarter Over Time
In the background, memories are continuously refined and consolidated — merging related facts, resolving contradictions, and strengthening important connections. Like how your brain processes memories during sleep.
Three separate conversations about your API design get merged into one coherent architectural summary — automatically.
One runtime. Many agents.
Shared intelligence.
Capabilities
Everything an agent needs to never lose the thread.
Living Context Window
No more aggressive compaction that throws away what matters. The context is a continuously adapted sliding window — always tuned to the current goal, always making room for what's relevant next.
Layered Memory with Procedures
Facts, decisions, goals, evidence, and files — each with its own lifecycle. Procedures get a dedicated repository so your agent always knows the right way to do things, not just what was done.
Multi-Agent Shared Memory
Multiple agents share the same runtime. Knowledge flows between them or stays isolated — per session, per actor, per team. Configurable via policies with built-in presets to get started fast.
Red-Line Guards
Enterprise-grade safety built in. Define procedures and policies that agents must follow. Red-line guards prevent costly mistakes — like deploying on a Friday or overwriting production data.
Decoupled Runtime
Purpose-built for OpenClaw, but the ElephantBroker runtime decouples agent-specific details. Memory or context management can run independently — in OpenClaw or any other agent harness.
Prometheus Monitoring
Out-of-the-box Prometheus-compatible metrics endpoints. Monitor your entire agent fleet's memory health with Grafana dashboards — industry-standard observability, zero custom tooling.
Enterprise Grade
Agents that follow
the rules you set.
In production, "smart" isn't enough — agents need to be safe. ElephantBroker embeds enterprise controls directly into the memory and context layer, so guardrails travel with the agent regardless of the task.
Procedural Memory
Dedicated storage for runbooks, workflows, and standard operating procedures. Agents don't just remember what happened — they remember how things should be done.
Policy Engine
Define rules that govern agent behavior. Policies control what can be shared, who can access what, and which actions require approval.
Red-Line Guards
Hard boundaries that agents cannot cross. Prevent destructive actions, enforce compliance requirements, and block operations that violate your organization's rules.
Actor Awareness
Every participant — human, supervisor, worker agent, or external service — is tracked with their own goals, permissions, and trust level.
Red-Line Guard in Action
Executing DROP TABLE users on production database...
Blocked. Policy no-destructive-prod-ops prevents DROP operations on production. Requires human approval.
Understood. Requesting approval from @admin before proceeding.
Procedural Memory
Deploy the new API version.
Following stored deployment procedure: (1) Run test suite, (2) Build Docker image, (3) Deploy to staging, (4) Run smoke tests, (5) Promote to production. Currently on step 1...
Live Monitoring Dashboard
Prometheus + GrafanaStored Memories by Type
Context Window
Utilization
Guard Blocks
Blocked this week
The Difference
What changes when your agent remembers.
Without ElephantBroker
What database are we using?
I don't have that context. Could you remind me?
Deploy to staging
What's the staging URL? I don't see it in this conversation.
Fix the auth bug from last week
I'm not sure which bug you're referring to. Can you describe it?
With ElephantBroker
What database are we using?
PostgreSQL — you decided this in session 3 when we compared it against MongoDB for the user table.
Deploy to staging
Following the deployment procedure: running tests first, then building the Docker image. Staging URL is staging.acme.dev.
Fix the auth bug from last week
Found it — the JWT expiry was set to 5 min instead of 60. Here's the fix based on the auth module we built in session 12.
Architecture
OpenClaw native.
Universally portable.
Fully local.
ElephantBroker is a single cognitive runtime with a shared knowledge plane, policy layer, and scoring framework. Built natively for OpenClaw's sub-agent delegation model, but architected so that either memory or context management can run independently in any agent harness — thanks to the decoupled runtime layer.
OpenClaw Plugin Surfaces
Native integration with OpenClaw's agent framework — Memory Plugin and Context Plugin as first-class product interfaces. Purpose-built for OpenClaw's sub-agent delegation model.
ElephantBroker Runtime
The decoupled cognitive core. 15+ modules including ActorRegistry, GoalManager, WorkingSetManager, CompactionEngine, and RedLineGuards. Runs in OpenClaw or any other agent harness — memory and context can operate independently.
Knowledge Plane
The shared knowledge substrate — graph entities, vectorized data points, enrichment pipelines, and ontologies. Multiple agents connect to the same plane, with policy-controlled isolation and sharing.
Infrastructure
Pluggable storage backends — vector stores, graph databases, relational databases, caches. Prometheus-compatible metrics endpoints for Grafana monitoring. The entire stack runs locally for maximum privacy and performance.
Your data never
leaves your servers.
The entire ElephantBroker stack — runtime, knowledge plane, vector stores, graph databases, and monitoring — runs on your own infrastructure. No cloud dependencies. No data leaving your network. No third-party access to your agents' memories.
For regulated industries, sensitive IP, or teams that simply want full control — local deployment isn't just an option, it's the default architecture.
Full Local Deployment
Every component runs on-premise. Docker Compose for development, Kubernetes-ready for production. No external API calls required.
Zero Data Exfiltration
Agent memories, context, procedures, and policies stay within your network boundary. Ideal for healthcare, finance, defense, and proprietary R&D.
Maximum Performance
No network round-trips to external services. Sub-millisecond memory retrieval. Context assembly happens at the speed of local compute.
Industry-Standard Monitoring
Prometheus metrics endpoints expose memory health, context utilization, guard activations, and more. Connect your existing Grafana dashboards — no proprietary tooling.
Intelligent Scoring
Not all memories
are created equal.
Every piece of stored knowledge competes for a spot in the living context window. Multiple signals are weighed together to surface exactly what matters — and different agent profiles (coding, research, management) tune these weights differently.
Think of it like a newsroom editor deciding which stories make the front page. Space is limited, so only the most important and relevant information gets through.
Use Cases
Built for every agent that needs to remember.
Start with a preset for common roles, or design a completely custom setup. From a single personal assistant to a fleet of enterprise agents with complex hierarchies — ElephantBroker adapts to your structure.
Ready-to-use Presets
Software Engineer
PresetRemembers codebase patterns, tracks open issues across sessions, follows deployment procedures, and recalls architectural decisions made weeks ago.
Research Analyst
PresetBuilds a growing knowledge base, cross-references sources, shares findings with other agents, and never re-reads a paper it already analyzed.
Project Manager
PresetTracks delegated tasks, maintains sprint goals, enforces organizational policies, and ensures nothing falls through the cracks across weeks of work.
Personal Assistant
PresetKnows your preferences, remembers appointments, tracks health notes, gift ideas, and personal commitments — like a chief of staff who never forgets.
Marketing Strategist
PresetTracks campaign performance over time, remembers brand guidelines, maintains audience insights, and builds on past creative decisions instead of starting fresh.
Domain Expert Advisor
PresetAccumulates deep expertise in any field — legal, medical, financial. Remembers case history, precedents, and evolving regulations across every consultation.
Custom Configurations
Organizational Hierarchy
CEO agent delegates to VP agents, who delegate to team leads. Policies cascade down. Each level sees only what it should. Memory flows up as summaries.
Social & Collaborative
Agents that work like friends — sharing context casually, remembering shared experiences, building on each other's ideas without rigid structure.
Isolated Expert Pool
Multiple specialist agents (legal, finance, engineering) with isolated memories. A coordinator agent queries each one and synthesizes answers.
Build Your Own
Define custom memory sharing rules, context policies, guard rails, and scoring weights. Every setup is configurable through the policy engine.
Stop re-explaining.
Start building.
ElephantBroker is in active development. Get early access to the runtime, presets, and documentation — and help shape what comes next.