#1 on BEAM at 128K tokens

Harny turns on your AI.

From AI API key to a working agent. AI that forgets its users is finished. Harny gives every user your agent serves a memory that keeps up when facts change, the accounts they already use, and a record of every action. One key. Nothing else to build.

Set up in 2 minutes with your agent

Your agent. Your model. Your users, remembered.

No published memory system beats Harny at 128K tokens on BEAM. 85.1, the highest published score at this size, on everyday models. BEAM is the public benchmark for long-term memory in agents.

85.1128K tokens

98% fewer tokens than handing your agent raw tool output. 4× cheaper per answer. How we measured

Works with 20+ agent frameworks.

85.1 on BEAM at 128K tokens.

Private by design: every user's memory is scoped to that user.

Quickstart

  1. Get API keyFull quickstart
ConsoleAPI keys
App key••••••••••••••••Copy

Store it in your secret store as HARNY_APP_KEY.

export HARNY_APP_KEY="…"
Create key

Benchmark results

BEAM accuracy at 128K tokens, ranked #1 among published results.

85.1BEAM, 128K tokens
98%fewer tokens

85.1 on BEAM at 128K tokens.

508 of 600 on DolphinBench, in our internal runs.

The benchmarks

Four API endpoints

Ingest timestamped text, check its status, delete it when required, and recall current facts with evidence.

Use the memory API alone, or switch on actions, permissions and connected accounts with the same key.

API reference

POST /v1/memory/ingestRaw text plus its original timestamp
GET /v1/memory/ingest/{ingestionId}Poll until status is completed
DELETE /v1/memory/ingest/{ingestionId}Remove the send's data points and every memory that rests only on them
POST /v1/memory/recallReturn ranked memory for your app or your model

What Harny handles

Backdatable ingestion

Backfill existing data with its original timestamps.

Memory per user

Each recall names the user. It returns what that person's memory holds, and nothing from anyone else's.

Dated evidence

Every result names the data point behind it and the date that data point happened.

Per-call usage

The console records each ingest and recall, and the usage route returns the month's count.

Long jobs

One agent, one folder, one hour. Hand it a job in one sentence and it works through the files.

Long jobs

Token ceiling per call

8,000 by default, or the budget you name, up to your plan's ceiling. Each recall reports the tokens it used.

Limits

Current-state tracking

Harny tracks current and earlier values, with their dates and sources.

How it works

Recall for one person

Name the user making the request. Harny returns the memory that user holds, with dated sources.

Access control

The best published BEAM score at 128K

BEAM is the public benchmark for long-term memory in agents, and Harny leads it at 128K tokens.

How to benchmark agent memory

Deployment options

Managed cloud

Harny runs the service. You call the API and nothing else changes.

Using MCP?

The address beside this text is Harny's MCP endpoint. Claude, Cursor, VS Code and other clients that support remote MCP servers connect to it with your app key.

https://mcp.harny.ai

MCP docs

Plans

Billed per person served, never per call. The first 500 people are free for good.

Free

$0 / month

For starting with your own app and growing at your own pace.

  • The first 500 monthly-active end users, free for good
  • One active app
  • More apps wait, paused, with nothing lost
Get API key

Paid

$0.30 per user above 500

For an app in production, with every app active.

  • The first 500 monthly-active end users stay free
  • Billed for monthly-active end users above 500
Get API key

Enterprise

Custom

Higher volume, priced against your usage and your terms.

  • Volume pricing
  • A direct line to the team
Talk to us

Full pricing

Frequently asked questions

01What do I send?

Text plus a timestamp: emails, meeting transcripts, chat messages, support threads, CRM notes, documents. There is no schema to define, no embedding model to choose and no chunking pipeline to build.

02What comes back?

Recall returns ranked memory for your model. The response carries asOf, usedEvidenceTokens and results. Each result is one memory with its content, its artifact and the dated data points behind it.

03How is this different from search or RAG?

Search and RAG return text that looks like the question. Harny returns what is true now, what it replaced and where each value came from. Ask about the Acme budget and you get 40k, with the date it changed and the email that changed it.

04Whose memory is it?

Each user has their own memory inside your app. A call names the user, and Harny answers from that user's memory only.

05How much does it cost?

The first 500 monthly-active end users in your active app are free for good. You pay only for users above 500, and a user counts in a month when they used your app that month. Every plan is on the pricing page.

06How do I get access?

Sign up for an account and create an app key in the console. New accounts start on Free.

07Need Harny in your own environment?

Talk to us.

Go on. Get Harny.

The first 500 monthly-active end users are free for good. Create an app key and make your first call in minutes.