Memory API FAQ

What is the base URL for the API?

The base URL is https://api.harny.ai. The Memory API is under https://api.harny.ai/v1/memory: send Authorization: Bearer $HARNY_APP_KEY and X-Harny-User: <user> on every call. The MCP server is at https://api.harny.ai/v1/mcp/mcp, and the machine-readable description of the API is at https://api.harny.ai/openapi.json.

Can I use Harny only as a memory API?

Yes. The Memory API (ingest, status, delete, recall) works on its own with one app key. Actions, permissions and connected accounts are optional, and they use the same key.

What is the most tokens a recall returns?

You set the ceiling on every recall. level picks 8,000 (the default), 16,000, 24,000 or 32,000 tokens, and maxTokens names any budget from 256 tokens. A plan can hold a lower ceiling: on the Free plan a recall returns at most 10,000 tokens. The answer reports usedEvidenceTokens, the tokens it used. See Limits.

How is Harny billed?

Harny bills per person your app served in a month, never per call or per token. The first 500 monthly-active end users are free for good. See Pricing.

Can I unit-test Harny?

Yes. Every route answers JSON with fixed field names, and every refusal uses one body, {code, status, debugMessage, data}, so a test asserts on fields and codes. GET /v1/memory/ingest/{ingestionId}/explain tells you what each item became and why.

Is Harny a black box?

No. Recall returns evidence, not an answer: each result names the data point and the date it happened. Your model writes the answer, and you keep control of the wording.

How do I audit what an agent did?

Every agent action is written to a numbered record, and each entry carries a hash of the one before. Export it with GET /v1/actions/export as JSON, or as CSV with format=csv. Add verify=true and Harny reports whether the sequence has gaps and whether the chain holds. The Requests screen of the console and GET /v1/memory/requests list each accepted recall, ingest, delete and recorded action with its id, and GET /v1/memory/trace/lineage/{rootId} gives the lineage of one root request.

Which models does Harny work with?

Any model works with the Memory API: recall returns ranked evidence as JSON, and your own model, on your key, writes the answer. The agent harness runs on your own key with the providers listed in Models and keys. Changing the model is one setting on your side.

How do I delete or export?

Delete one data point with DELETE /v1/memory/data-points/{sourceId}, or erase a person across every store with DELETE /v1/users/{user_id}/data; erasure finishes within 30 days. Export a person's memory as JSON with GET /v1/memory/users/{identity}/export, and the action record with GET /v1/actions/export.

How do I correct a data point?

Send it again with the same id and the new content. Harny replaces the data point and the memories built on the old text.

Do I need a memory API, or is a database or a markdown file enough?

For one person and one session, a file or a database table is enough. A memory API earns its place when your app serves many people and facts change: it keeps each person's memory apart, replaces a fact when it changes, returns dated evidence under a token ceiling you set, and erases one person across every store.

Does Harny use my data to train AI models?

No. Harny does not use your data to train general AI models. See Security.

Where is my data stored?

Data stored in the US. Model providers follow the app's model setting.

Is there a status page or an uptime commitment?

The Status page shows whether the API is up now. Free and Paid carry no uptime commitment. Enterprise terms are set by agreement.

Can I run Harny in my own cloud?

Harny runs as a managed service today. To run Harny in your own cloud, write to support@harny.ai with Enterprise in the subject.

Which languages and frameworks does Harny work with?

The Memory API is plain HTTP, so any language works; the docs show cURL, Python and TypeScript. Integration pages cover LangChain, LangGraph, LlamaIndex, CrewAI, the Vercel AI SDK, Microsoft Agent Framework, Google ADK, the OpenAI Agents SDK, Pydantic AI, Mastra, Strands and more at Integrations. The MCP server is at https://api.harny.ai/v1/mcp/mcp.

How long is a version supported?

/v1 stays stable. Breaking changes come in a new version. A version is retired with at least 6 months' notice, by email to the account owner and on the changelog.

Who builds Harny, and how do I get help?

Harny is built by Vexta Labs Inc., Burnaby, British Columbia, Canada. Write to support@harny.ai; we reply within one business day.

How are Harny's security controls organised?

Harny lists its security controls under the five category names of SOC 2's Trust Services Criteria: Security (one sign-in layer, API keys stored as one-way hashes and revocable at any time, every read scoped to the calling app, and the app's acting settings enforced on Harny's servers), Availability (rate limits and daily quotas that keep counting when a supporting service is down), Processing Integrity (a numbered, hash-chained action record you export and verify, and one error format on every route), Confidentiality (credentials encrypted before storage, credential-like strings removed before a model sees content, and fetched source copies deleted on a schedule) and Privacy (delete any stored item with the memories that rest only on it, and a retention schedule for children's data). See Security.

Is Harny the lfnovo/harny GitHub project?

No. Harny (harny.ai) is built by Vexta Labs Inc. It is not related to the GitHub repository lfnovo/harny, which is a different project.

Next

• Limits
• API reference
• Errors