Give your agent memory in four calls

Most agents forget everything between runs. agentics.you gives them a memory they can write to and recall from over plain HTTPS, from curl. Every write leaves a receipt, so you can prove later what your agent remembered. The same four calls are in TypeScript and Python on Save and recall your first memory. New here? Create an account first.

1. Create a key

Sign up, open API keys in the console, and choose Create key. A new workspace can create its first key right away: its default namespace is set up on first load. The key starts ak_, is shown once, and can read and write the default namespace from its first call. Keep it in an environment variable, never in code.

export AGENTICS_API_KEY="ak_..."

A call with no key answers HTTP 401:

{"error":{"code":"unauthorized","message":"Sign in, or send an org API key as Authorization: Bearer ak_... (create one with POST /api/api-keys)."}}

This key is for /v1. It does not work on /mcp. Connecting Claude, ChatGPT, or Cursor is a different sign-in: Connect your AI.

2. Remember something

Send the memory as content. Leave out namespace to use the default one.

curl -sS -D - https://agentics.you/v1/memories \
  -H "Authorization: Bearer $AGENTICS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"content":"The customer prefers email over phone."}'

The reply is HTTP 202 with a job_id. The memory is stored, and a job makes it searchable:

{
  "job_id": "5b0d7c1e-2f4a-4c8e-9a61-3d2f8b7e4a10",
  "memory_id": "9e6a2c34-7b1d-4f0e-8c5a-1a2b3c4d5e6f",
  "duplicate": false,
  "stage": "queued"
}

Read GET /v1/ingest/{job_id} until stage is done. Sending the same content to the same namespace again returns HTTP 200 with duplicate true and no new job.

curl -sS "https://agentics.you/v1/ingest/$JOB_ID" \
  -H "Authorization: Bearer $AGENTICS_API_KEY"

{
  "job_id": "5b0d7c1e-2f4a-4c8e-9a61-3d2f8b7e4a10",
  "memory_id": "9e6a2c34-7b1d-4f0e-8c5a-1a2b3c4d5e6f",
  "stage": "done",
  "failed_stage": null,
  "failure_reason": null,
  "terminal": true
}

A failed job names failed_stage and failure_reason. POST /v1/ingest/{job_id}/retry runs it again.

3. Recall it in your own words

curl -sS https://agentics.you/v1/memories/search \
  -H "Authorization: Bearer $AGENTICS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query":"how does the customer like to be contacted?"}'

Search combines keyword and meaning matching. Each hit comes back with a score, why it ranked, and the receipt_hash of its write. mode is hybrid. If the meaning step cannot run, mode is keyword and degraded says why (dense_quota when the day's query-embedding budget is spent). The reply says so. It does not switch in silence. Every search writes a recall receipt of its own.

{
  "query": "how does the customer like to be contacted?",
  "mode": "hybrid",
  "count": 1,
  "hits": [
    {
      "id": "9e6a2c34-7b1d-4f0e-8c5a-1a2b3c4d5e6f",
      "content": "The customer prefers email over phone.",
      "receipt_hash": "3f9c1ea07b",
      "score": 1,
      "why": { "arms": ["dense", "keyword"] }
    }
  ]
}

4. Check the receipt

Anyone can check a receipt without an account. Open https://agentics.you/verify/?hash= followed by the hash, or paste the hash on Check a receipt. From code:

curl -sS "https://agentics.you/api/proof/verify?hash=$RECEIPT_HASH"

{
  "verified": false,
  "receipt_recomputed": true,
  "chain_ok": true,
  "merkle_ok": false,
  "anchor_ok": false
}

That is a new receipt: its hash recomputes and it links to the receipt before it, and it is not in a batch yet. Once its batch is anchored, merkle_ok, anchor_ok, and verified turn true. The check never returns the memory's content. The receipt holds a digest of it.

Packages that call the same API

pip install agentics-you (PyPI, 1.2.2) and npm install @agenticstechnologies/sdk (1.2.0) call these /v1 routes with an ak_ key. The Python import is agentics_you, and AgenticsClient reads AGENTICS_API_KEY. agentics on PyPI is a different publisher. The MCP package @agenticstechnologies/mcp-server is not this API. It forwards a local client to https://agentics.you/mcp with an agk_mcp_ key.

Using Claude, ChatGPT, or Cursor instead?

Connect them to the same memory over MCP at https://agentics.you/mcp. One page per client: Claude, Claude Code, Codex, Gemini CLI, ChatGPT, Cursor, VS Code, Copilot CLI, Windsurf, Grok, or any MCP client.

What it costs

Shared memory is free to start, with no card. Paid plans are priced on actions, never per agent. See Pricing.

Start free