The pluggable runtime for AI agents.
On every turn, elephant.broker decides what your agents see. It assembles context from your tools, teammates and past sessions, drops what is stale or off-task, flags sources that disagree, and places the right procedure where the model pays attention.
Impact on your agents using elephant.broker
+36%
better accuracy
stale-state errors fell from 47% to 11%
+21%
more tasks completed
71.0% to 91.6% on the same 50 tasks
40–80%
less context per task
only verified, relevant knowledge reaches the window
60%
faster task runtime
60.2% less wall-clock time, same workflow
40%
lower AI spend
same model; savings compound on long-running work
71%
fewer tasks deferred
71% fewer non-completions, with 62.7% fewer tokens
Agents forget what your company already knows.
Context grows, accuracy falls
As conversations grow, models follow instructions with less reliable accuracy; by turn three, accuracy fell 17% across all 14 models tested.
Procedures don't stick
Instructions fade each turn; all fourteen tested models became less reliable as conversations continued.
Sessions from zero
Every agent starts from zero, unaware of what your tools, teammates, and other agents already learned across sessions and workflows.
Sources disagree
When two sources contradict each other, agents choose whichever memory ranks higher that turn, delete the conflicting fact, and cannot explain why.
Context failure
Models worsen as context fills, even with the right facts present. Reducing context increased completion from 71% to 92% using 63% fewer tokens.
We don't host your agents. We change what they see, every turn.
Your harness
Claude Code · Codex · OpenClaw · Hermes · Cursor · your own harness
elephant.broker
inside the loop · every turn · your model turn stays unchanged
Your business
your systems · your data · your decisions
How a turn flows through the runtime
Scroll the strip sideways to follow the whole path.
Agent turn
messages + tool calls
elephant.broker runtime
- Retrieve
- Rerank
- Working Set
- Assemble
- Policy enforcement
Your LLM
any model, unchanged
Every turn recalls from, and writes back to, one graph and vector store, gateway-isolated per tenant.
Plugs into any harness
Connect through the MCP server, native plugins or a REST API. Elephant.Broker changes what the model sees on every turn; your model call remains yours.
No agent hosting, no rip-and-replace and no lock-in.
- MCP server
- Native plugins
- REST API
Every turn: assemble, inject, better accuracy.
The runtime wraps the model turn you already own, then writes the outcome back for the next turn.
Pre-turn hook
Intercepts the turn before the model sees it.
Assemble
Pulls task-relevant context across tools, teammates and sessions.
Rank & prune
Keeps current, verified context and drops stale or duplicate facts.
Inject
Places the working set where the model pays attention.
Model turn
Your model, your harness, unchanged.
Post-turn hook
Captures decisions, corrections and outcomes.
Write back
Makes the result available to the next turn.
Safe, private, and yours
Security, privacy and data ownership are built in with zero credential takeover.
Your credentials never leave your side
We hold no connectors and no keys. Elephant.Broker keeps pointers to your systems, and your agents use the access you already gave them. Nothing moves.
Hardened deployments
Single-tenant by default. Private networking, encryption in transit and at rest, SSO/SCIM, scoped keys and a signed, append-only audit log.
Zero data retention. No training on your data.
Your memory is never used to train anyone's model. Zero-retention pass-through to model providers, with retention windows you set per memory class.
Cloud, private cloud or fully on-premise
EU regions and sovereign hosting, or self-hosted inside your own VPC or air-gapped network. Deploy where your policy says it has to live.
Everything the runtime does, in one layer.
Better context, every turn
Assembled from wide company context, pruned and placed where the model pays attention.
Procedures placed per turn
Your ways of working become versioned procedures, placed for the task at hand.
Disagreement flagged
When two sources conflict, both are placed and weighed by proof.
Tamper-evident audit & logs
Every read, placement, write-back and approval is recorded against the actions taken.
Pluggable memory
Bring the memory you already run, or use ours. The layer stays fully pluggable.
Procedure & workflow capture
Your ways of working become versioned procedures agents actually follow.
Spend dashboards & visibility
See what every agent, task and procedure costs, and cap it.
Guardrails & approvals
Red lines are blocked before execution; humans approve where it matters.
More room on every turn
A smaller working set leaves headroom for longer tasks and more agents inside the same window. The runtime sends 40–80% less per turn, and your per-task numbers land on your dashboard from day one.
The architectural difference: pointer-first and runtime-native.
Scroll the table sideways to see every column.
| Versus | Others | EB. |
|---|---|---|
| Context warehousesModus · ServiceNow · Hyland | Connector-first: ingest and copy enterprise data into their store; tied to their platform. | Pointer-first: nothing moves; the decision is made at runtime, per turn, in any harness. |
| Memory libraries and cross-tool memory serversMem0 · Cognee · Supermemory · Zep · Letta | Store and retrieve facts; auto-resolve or delete on contradiction. | Decide what enters the window this turn and what stays out; flag disagreement and weigh by proof. |
| Harness-native memoryClaude Code Auto Memory · Codex memory · managed-agent memory | One user, one project, one vendor; static rules files. | Shared across agents, tools and teams; the relevant procedure is placed per turn. |
The team
The team behind elephant.broker
Sarunas Legeckas
CEO & Co-founder
- Operator
- Ex-N26 GM across 18 EU markets.
- Go-to-market
- Runs enterprise pilots, partnerships & rollout.
Alexandru Lupascu
CTO & Co-founder
- Engineering
- Ex-CTO & co-founder @ OMNIA Protocol. Patent EP3883204B1.
- AI runtime
- Ships the agent runtime, integrations & audit engine.
Cristian Lupascu, PhD
CPO & Co-founder
- AI research
- Ex-CEO & co-founder @ OMNIA Protocol. PhD in privacy-enhancing technologies.
- Product & Security
- Designs the product features & security.
Simas Simanauskas
Founding partner, GTM
- Commercial
- Ex-Chief Commercial Officer @ ConnectPay. Drove 18x client growth.
- Go-to-market
- Leads sales, design partners & key accounts.
Run accurate agents in production with a record of every action.
We're opening a small number of design-partner pilots. Pick the door that matches how you evaluate.
For business operators
Pick the pipeline together and agree the baseline.
A 30-minute discovery, then a sandboxed pilot: single-tenant, your data, your numbers at the end.
For engineers
Architecture and security first. Then a sandbox.
One page for your CTO and security reviewer. Then inspect how the runtime retrieves, guards, and logs every action. Your model, your harness, unchanged.