AmpUp Memory

AmpUp Memory is the memory for agents that gets better with use. A conversation is a session; each task in it asks for advice and gets one skill for that kind of task: a note learned from earlier rewarded runs or, before any exist, a procedure compiled from your documents. When the session ends, its transcript and outcome are judged or rewarded, and a rewarded session teaches the next one. Underneath is a cited memory: notes, documents and conversations become searchable passages and facts, each fact quoting its source and visible only to people who may read every source it cites. Everything is stored as Apache Iceberg tables in a catalog and bucket you run.

GitHub

Source code, Apache-2.0. Issues, discussions and contributions.

Why it learns

How the learning loop works, and the benchmarks behind it.

Connect your agents

Claude Code

A plugin with hooks: every session asks for advice and records how it went.

Codex, Cursor, Gemini CLI, Copilot

The same loop through ampup-memory hooks install.

Any MCP client

Search, remember and get advice over MCP at /mcp.

Your own agent

Python and TypeScript clients, or plain HTTP.

Run it locally

Build from source (you need rustup and protoc), then start a local server and point Claude Code at it:

git clone https://github.com/A79-ai/ampup-memory && cd ampup-memory
cargo install --locked --path .        # the ampup-memory server and CLI

OPENROUTER_API_KEY=... AMPUP_MEMORY_EXTRACT_GOAL_BLIND=1 ampup-memory serve --local   # http://127.0.0.1:8080
claude mcp add --transport http ampup-memory http://localhost:8080/mcp

--local needs no Docker, catalog service or sign-in, and binds to 127.0.0.1. Without a model key it stores and searches but learns nothing; any OpenAI-compatible, Anthropic, Bedrock or Vertex AI endpoint, or a local Ollama, works instead of OpenRouter (configuration). The quickstart walks through a first note, search and piece of advice, including fully offline on a laptop.

From code

pip install ampup-memory        # Python 3.10+
npm install ampup-memory        # Node 20+

from ampup_memory import AmpUpMemory

mem = AmpUpMemory("http://127.0.0.1:8080", None)
mem.remember("note:deploy", "Deploy billing with ./deploy.sh billing.")
print(mem.query({"find": "passages", "text": "deploy billing"})["results"])

run = mem.session("deploy the billing service")  # advice per task, rewarded at the end
print(run.opening["advice"])
run.end("deployed", reward=1)

The clients wrap the HTTP routes: /remember, /search, /query, /advice, /sessions, /sessions/end, /runs, /runs/reward and /mcp (all routes).

Your team’s hosted memory

If AmpUp runs a memory for your company, your admin gets an invite email to set up the org; see Set up your org and Invite people.

Guides

Concepts

Scopes, episodes, facts, branches, sessions and advice.

Advice

How notes are learned, judged and gated.

Schemas and queries

Typed memory, extraction and search ranking.

Self-hosting

Iceberg REST (Polaris), BigLake, Snowflake, Databricks; GCS, S3, Azure.

CLI

The ampup-memory and ampup-memory-client commands.

Security model

Identity, ACLs, API keys, audit, deletion and retention.

Using it

Running it

Your org