Persistent memory for
AI agents.
No database to manage.

One HTTP API for persistent memory, shared namespaces, TTLs, and tenant isolation, whether you run one agent or a whole team. No vector database. No embedding pipeline. No memory infrastructure to build yourself.

agentram api
# Store a memory
POST api.agentram.dev/memory
agent_id: "agent-01" key: "user_lang" value: "French"
{ "success": true, "credits_remaining": 999 }

# Search across all memories
GET api.agentram.dev/memory/search?agent_id=agent-01&q=language
{ "data": [{ "key": "user_lang", "value": "French" }] }

# Share memory across multiple agents
POST api.agentram.dev/memory/shared
namespace_key: "ns_team01" key: "company_tone"
{ "success": true }
1,000 free credits on signup
24/7 live uptime monitor
1¢ per 100 operations
0 infra to manage
// the amnesia test // no signup

Watch one agent forget.
And one remember.

Tell both agents a few things about yourself. Start a new session, the way your users do every day. Then ask what you said.

Plain agent
no memory layer
Agent + AgentRAM
persistent memory
memory: empty
Illustrative agent replies. The API calls shown are AgentRAM's real request and response shapes.

Set up in minutes.
Ship the same day.

Three endpoints. No infrastructure. No setup headaches. Your agents remember things now.

01.
Register and get your key

Send your email and we send back an API key with 1,000 free credits. No credit card required to start.

POST /register
→ key sent to your email
02.
Write what your agent knows

Store anything your agent needs to remember. User preferences, task state, conversation context. Each write costs one credit.

POST /memory
agent_id · key · value
03.
Read it back in any session

Retrieve any memory later. Your agent picks up where it left off, with the facts and state you saved.

GET /memory?agent_id=...&key=...
→ value: "..."
04.
Monitor your balance

Check remaining credits at any time. Top up when needed. This endpoint never deducts a credit.

GET /credits
→ credits: 847

Everything your agents
need to remember.

No unnecessary context bloat. No graph configuration. No embedding pipeline. Memory that works the first time and keeps working.

One call to remember

Store and retrieve memory with a single HTTP request. No retrieval strategies to configure, no embeddings to tune, no query language to learn. Your agent asks by key and gets its answer.

Fully isolated namespaces

Each API key operates in a private namespace. No agent can read another agent's memories. No shared state unless you choose it.

Search across memories

Find any stored memory by text in a single call. No embeddings to configure, no vector index to provision, no retrieval pipeline to maintain.

Updates, not duplicates

Writing to an existing key updates the value in place. No stale memories building up. No cleanup logic to write yourself.

History when you need it

Some facts change and you need to know what they used to be. Assert a value instead of overwriting and you get the whole chain: what it was, who changed it, and when. If two agents disagree, a read says so rather than quietly picking one. Temporal memory →

Memory expiry built in

Set a TTL on any memory and it disappears automatically when it is no longer relevant. No cron jobs. No manual pruning. Clean context by default.

Shared memory for agent teams

Create a shared namespace and let multiple agents read and write to the same memory pool. Built for multi-agent workflows from the ground up, not bolted on.

Works with any framework

Plain HTTP. If your agent can make a web request, it works. No SDK required. No lock-in to any orchestration tool, agent runtime, or cloud provider.

Predictable cost, always

One operation costs one credit. Not one token. Not one episode that multiplies with conversation length. You know what you are paying before you pay it.

You never pay for a failed request

Credits deduct only on success. If a request fails for any reason, your credit returns automatically. No billing surprises from operations that did not complete.

Plug memory into the
tools you already use.

AgentRAM ships an MCP server, so any MCP-compatible agent gets persistent memory from one config block. Or drop in the SDK. No framework rewrite, no lock-in.

MCP Any MCP client Claude Code via MCP Codex via MCP OpenAI Agents SDK via MCP

More agents (OpenClaw, Hermes) are on the way. Building on something else? It is just HTTP, so any framework works.

Find out if you
actually need this.

An honest self-check, and the reasoning behind it. No signup, no hard sell. If you do not need a memory layer yet, it will tell you so.

Take the 2-minute assessment →

Ten quick questions that tell you whether you need a memory layer, which kind, and what to do next. It says "not yet" when that is the honest answer. No signup.

Why agents need memory →

The benefits, one by one: continuity across sessions, personalization, leaner and cheaper prompts, reusing what already worked, and coordinating multiple agents.

Pay for what you use.
Nothing more.

One operation, one credit. No per-token billing that explodes as your agents get smarter. Credits never expire.

Scale
$40
600,000 operations = $0.067 per 1,000
  • Read, write and delete
  • Full API access
  • Priority support
  • Lowest per-operation cost
Get your key →
Starter
$5
50,000 operations = $0.10 per 1,000
  • Read, write and delete
  • Full API access
  • Email support
Get your key →

Your agents are ready
to start remembering.

1,000 free operations on signup. No credit card required.

Get your API key →

© 2026 AgentRAM. All rights reserved.