Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
25 commits
Select commit Hold shift + click to select a range
b27ce7f
Add the LongMemEval harness under agent-memory-benchmark/.
ichko Aug 31, 2026
d13fe8a
Present Redis Agent Memory, V0, and the LongMemEval harness together.
ichko Aug 31, 2026
6fbc971
Give each memory adapter its own module.
ichko Aug 31, 2026
a4aa659
Drop the Mastra OM and Emergence Fast placeholders.
ichko Aug 31, 2026
c7efde1
Add a Redis Agent Memory store for the LongMemEval harness.
ichko Sep 1, 2026
cc98430
Fix adapter SDK mismatches and runner isolation bugs.
ichko Sep 1, 2026
0572db9
Fix judge formatting, bounded Redis cleanup, and Bedrock reset.
ichko Sep 1, 2026
5076c0b
Wait for stable extraction and keep failed cleanup retryable.
ichko Sep 2, 2026
359fcfd
Lead the landing copy and store registry with Redis Agent Memory.
ichko Sep 2, 2026
48fce67
Page memory listings and pass session dates into adapters.
ichko Sep 2, 2026
c8adb99
List memories without dummy search and keep digit IDs as strings.
ichko Sep 2, 2026
eafa48e
improve readmes and some docs
tylerhutcherson Sep 2, 2026
d217f65
Stream LongMemEval downloads and fail incomplete extraction waits.
ichko Sep 4, 2026
0fa60ad
Merge branch 'add-agent-memory-benchmark' of https://github.com/ichko…
ichko Sep 4, 2026
70bc8f5
Surface Redis long-term memory delete failures.
ichko Sep 4, 2026
f36ede1
Lengthen extraction wait and put session dates into Mem0 context.
ichko Sep 4, 2026
50e77c2
Match provider SDK argument names and keep resume duration.
ichko Sep 4, 2026
82d3af5
Skip inherited extraction waits and ingest session dates.
ichko Sep 4, 2026
8cc8460
Poll Zep and Bedrock until extraction is stable.
ichko Sep 4, 2026
7cd363c
Handle empty Graphiti edge groups.
ichko Sep 4, 2026
d7463e9
Fix harness lint so GitHub CI can pass.
ichko Sep 4, 2026
af09623
Scope Vertex list and reset to the example user.
ichko Sep 4, 2026
92b7be7
Delete Oracle ingest threads on reset.
ichko Sep 4, 2026
8959016
Restore the original LongMemEval judge, answer prompt, and hypothesis…
ichko Sep 15, 2026
e25f7cc
Lead the landing README with the Redis Agent Memory product.
ichko Sep 15, 2026
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
33 changes: 33 additions & 0 deletions .github/workflows/agent-memory-benchmark.yml
Original file line number Diff line number Diff line change
@@ -0,0 +1,33 @@
name: Agent Memory Benchmark

on:
push:
pull_request:

permissions:
contents: read

jobs:
test:
runs-on: ubuntu-latest
defaults:
run:
working-directory: agent-memory-benchmark
steps:
- uses: actions/checkout@v4
- uses: astral-sh/setup-uv@v6
with:
enable-cache: true
cache-dependency-glob: agent-memory-benchmark/uv.lock
- name: Install Python
run: uv python install 3.11
- name: Install dependencies
run: uv sync --locked --group dev
- name: Lint
run: uv run ruff check .
- name: Check formatting
run: uv run ruff format --check .
- name: Test
run: uv run pytest
- name: Build package
run: uv build
16 changes: 10 additions & 6 deletions AGENTS.md
Original file line number Diff line number Diff line change
@@ -1,16 +1,20 @@
# AGENTS.md

The project source, build tooling, tests, and docs live in [`V0/`](./V0/).
The repository root is a thin landing layer — see [`README.md`](./README.md). Two
projects live below it. Each has its own tooling, tests, and docs. Run build,
lint, and test commands from inside the project directory, not from the root.

**Start there:** read [`V0/AGENTS.md`](./V0/AGENTS.md) for the full agent and
development instructions. Run build, lint, and test commands from inside `V0/`
(for example `cd V0 && make test`).
| Directory | Project | Instructions |
|-----------|---------|--------------|
| [`V0/`](./V0/) | Agent Memory Server, the pivotal research foundation preserved as a V0 artifact | [`V0/AGENTS.md`](./V0/AGENTS.md) (and [`V0/CLAUDE.md`](./V0/CLAUDE.md)) |
| [`agent-memory-benchmark/`](./agent-memory-benchmark/) | Open LongMemEval audit harness (`memory-bench`) | [`agent-memory-benchmark/README.md`](./agent-memory-benchmark/README.md) |

The repository root is a thin landing layer — see [`README.md`](./README.md).
Examples: `cd V0 && make test`, or
`cd agent-memory-benchmark && uv run pytest`.

## Looking for the supported production path?

This repository (`V0/`) is the open-source research foundation. Redis's official
The code in this repository is open-source research and evaluation work. Redis's official
managed path for production use is **[Redis Agent Memory in Redis Iris](https://redis.io/agent-memory/)**
— agent memory as a service, with a REST API, client libraries, secure API key
management, and automatic TTL-based lifecycle management. See the
Expand Down
26 changes: 22 additions & 4 deletions CLAUDE.md
Original file line number Diff line number Diff line change
@@ -1,5 +1,23 @@
# CLAUDE.md - Redis Agent Memory Server Project Context
# CLAUDE.md - Redis Agent Memory Project Context

This project lives in [`V0/`](./V0/). See [`V0/AGENTS.md`](./V0/AGENTS.md) for
full instructions (and [`V0/CLAUDE.md`](./V0/CLAUDE.md)). Run commands from
inside `V0/`.
Read [`AGENTS.md`](./AGENTS.md) first. It maps the two projects in this
repository: [`V0/`](./V0/) (the Agent Memory Server) and
[`agent-memory-benchmark/`](./agent-memory-benchmark/) (the LongMemEval
harness).

For the server, see [`V0/AGENTS.md`](./V0/AGENTS.md) and
[`V0/CLAUDE.md`](./V0/CLAUDE.md). For the harness, see
[`agent-memory-benchmark/README.md`](./agent-memory-benchmark/README.md).

Run commands from inside the project directory, not from the repository root.

Keep the repository narrative intact when changing landing or project context:

- `V0/` is intentionally preserved as the pivotal open research foundation that
informed Redis Agent Memory productization, not as the supported production
distribution.
- Redis Agent Memory in Redis Iris is the official product path; keep its
create-service and SDK/API quickstarts easy to find.
- The benchmark is an open audit and reproducibility surface, not a timeless
vendor ranking. Benchmark claims must name the provider configuration, split,
models, retrieval settings, artifacts, and run date.
63 changes: 35 additions & 28 deletions README.md
Original file line number Diff line number Diff line change
@@ -1,50 +1,57 @@
<div align=center>

# Redis Agent Memory

A memory layer that gives agents intelligent short-term memory and persistent context across conversations.
[Redis Agent Memory](https://redis.io/agent-memory/) is Redis's managed memory layer for AI agents: a persistent store that lets an agent recall facts, events, and preferences across sessions, without stuffing every token of history back into the prompt. It is part of [Redis Iris](https://redis.io/iris/), Redis's real-time context engine for agents.
Comment thread
tylerhutcherson marked this conversation as resolved.
Resolved

</div>
[Redis Agent Memory docs](https://redis.io/docs/latest/develop/ai/context-engine/agent-memory/) · [Benchmark harness](./agent-memory-benchmark/) · [Benchmark write-up](https://redis.github.io/redis-ai-research-public/longmemeval-agent-memory/)

## Redis Agent Memory in Redis Iris
## Getting started

[Redis Agent Memory in Redis Iris](https://redis.io/agent-memory/) is Redis’s official managed path for teams that want agent memory as a service, not another subsystem to build and operate themselves. [Redis Iris](https://redis.io/iris/) is the real-time context engine for agents, designed to deliver fresh, relevant context at runtime, and Redis Agent Memory is the part of Iris that makes context compound across turns, sessions, channels, and agents.
- Product overview: [Redis Iris](https://redis.io/iris/)
- Agent Memory overview: [Redis Agent Memory docs](https://redis.io/docs/latest/develop/ai/context-engine/agent-memory/)
- Redis Cloud service guide: [Redis Agent Memory on Redis Cloud](https://redis.io/docs/latest/operate/rc/context-engine/agent-memory/)

Redis Agent Memory in Iris gives you the Redis-managed experience: a persistent, structured memory layer for AI agents exposed through a REST API and client libraries, with dedicated endpoints, secure API key management, configurable memory schemas, and automatic TTL-based lifecycle management. The point is not just storage. It is to remove the custom memory infrastructure teams otherwise end up building around session handling, extraction, retrieval, and lifecycle management.
On Redis Cloud:

Redis Agent Memory uses a two-tier model. Session memory keeps the active conversation state, session history, and session-specific metadata close at hand, with configurable TTL control for retention. Long-term memory stores extracted facts and learned patterns from past interactions as text plus vector embeddings for semantic retrieval. As new events are written to working memory, Redis Agent Memory automatically extracts important information and promotes it to long-term memory in the background, so memory accumulates without slowing down the live agent loop.
1. [Create a database](https://redis.io/docs/latest/operate/rc/databases/create-database/).
2. [Create an Agent Memory service](https://redis.io/docs/latest/operate/rc/context-engine/agent-memory/create-service/).
3. Use the [Python SDK](https://redis.io/docs/latest/develop/ai/context-engine/agent-memory/python-sdk-quickstart/), [TypeScript SDK](https://redis.io/docs/latest/develop/ai/context-engine/agent-memory/typescript-sdk-quickstart/), or [REST API](https://redis.io/docs/latest/develop/ai/context-engine/agent-memory/rest-api-quickstart/) from your app.
4. Manage the service with the [Redis Cloud guide](https://redis.io/docs/latest/operate/rc/context-engine/agent-memory/).

That matters because Redis Iris is not just a memory feature in isolation. It is a broader context engine built to address the production problems agents actually hit: fragmented data, stale operational state, slow retrieval, and interactions that do not improve over time. Within that story, Redis Agent Memory is the compounding memory layer; [Redis Context Retriever](https://redis.io/context-retriever/) makes business data navigable; [Redis Data Integration](https://redis.io/data-integration/) keeps operational state fresh; and [Redis LangCache](https://redis.io/langcache/) helps repeated work stay inside the latency budget.
Implementation details: [API and SDK examples](https://redis.io/docs/latest/develop/ai/context-engine/agent-memory/) and the [API reference](https://redis.io/docs/latest/develop/ai/context-engine/agent-memory/api-reference/).

If you are evaluating the supported Redis path, these are the best places to start:
## How it works

- Product overview: [Redis Iris](https://redis.io/iris/)
- Agent Memory overview: [Redis Agent Memory docs](https://redis.io/docs/latest/develop/ai/context-engine/agent-memory/)
- Redis Cloud service guide: [Redis Agent Memory on Redis Cloud](https://redis.io/docs/latest/operate/rc/context-engine/agent-memory/)
Redis Agent Memory uses a two-tier model. Session memory keeps active conversation state close at hand, with configurable TTL. Long-term memory stores extracted facts from past sessions as text plus vector embeddings for later retrieval.

As new events land in session memory, Redis Agent Memory extracts what matters and promotes it to long-term memory in the background. Memory can accumulate without slowing the live agent loop.

Keeping full conversation history in the prompt does not scale. Context windows are bounded, long prompts are slow and expensive, and accuracy degrades as prompts grow. Agent memory lets an agent stay fast and cheap on every turn while still recalling what happened weeks ago.

A practical getting-started flow on Redis Cloud looks like this:
## Validated independently

- [Create a database](https://redis.io/docs/latest/operate/rc/databases/create-database/)
- [Create an Agent Memory service](https://redis.io/docs/latest/operate/rc/context-engine/agent-memory/create-service/)
- [Use the Agent Memory API](https://redis.io/docs/latest/operate/rc/context-engine/agent-memory/use-agent-memory/) from your application
- [View and manage your service](https://redis.io/docs/latest/operate/rc/context-engine/agent-memory/view-service/)
We ran a best-effort LongMemEval comparison across a sweep of memory systems, on a shared answer model and judge, to put those systems on more equal footing (Wu et al., ICLR 2025). Some vendors publish higher scores than we report. We were not able to reproduce those numbers under this protocol.

For implementation details and usage examples, see:
A separate research configuration that combined two retrieval strategies (Instruct extraction and Remis) reached **86.5%** task-averaged accuracy in our evaluation, at about $0.07 per session. Remis is **not** in Redis Agent Memory in Iris yet, and that configuration is **not** what this public harness runs. Do not treat 86.5% as a Redis Cloud product score.

- [API and SDK examples](https://redis.io/docs/latest/develop/ai/context-engine/agent-memory/api-examples/)
- [API reference](https://redis.io/docs/latest/develop/ai/context-engine/agent-memory/api-reference/)
The public harness in [`agent-memory-benchmark/`](./agent-memory-benchmark/) evaluates the shipping Redis Agent Memory REST API and other adapters. It is an audit surface, not a timeless ranking. A result is evidence about a specific provider version, split, models, retrieval settings, artifacts, and run date.

```bash
git clone https://github.com/redis/agent-memory-server.git
cd agent-memory-server/agent-memory-benchmark
```

Setup, protocol, and provider recipes: [`agent-memory-benchmark/README.md`](./agent-memory-benchmark/README.md).

## V0 — the open-source research foundation

[**`V0/`**](./V0/) contains the original Redis Agent Memory Server: an open-source reference implementation for agent memory with REST and MCP interfaces, working and long-term memory, configurable extraction strategies, and Redis-backed semantic search.
It serves as the research foundation and architectural starting point for Redis Agent Memory, but it is not the current supported production path.
[`V0/`](./V0/) is the original Redis Agent Memory Server. It informed the Iris product. We keep it here as an open research artifact, not as the supported production distribution. Start at [`V0/README.md`](./V0/README.md).

## Citing this work

Redis Applied AI Research, "How Redis Agent Memory Stacks Up: Benchmarking Memory Systems for AI Agents on LongMemEval," 2026.

- **Start here:** [`V0/README.md`](./V0/README.md)
- **Documentation:** [`V0/docs/`](./V0/docs/index.md)
- Build, test, and run everything from inside `V0/` (e.g. `cd V0 && make test`).
Also see [Building and evaluating long-term conversational memory](https://redis.github.io/redis-ai-research-public/longmemeval-agent-memory/).

## License

This project is licensed under the **Apache License 2.0** (Redis, Inc.). See
[`LICENSE`](./LICENSE) at the repository root.
Apache License 2.0 (Redis, Inc.). See [`LICENSE`](./LICENSE).
10 changes: 8 additions & 2 deletions V0/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -11,13 +11,19 @@ A memory layer for AI agents.


>
> This is the original reference implementation of Redis Agent Memory Server. It remains a useful reference and starting point for understanding agent memory, but **it is no longer actively maintained and is not the supported production path.**
> This is the pivotal open-source research foundation behind Redis Agent Memory.
> Its two-tier architecture, extraction strategies, retrieval, and lifecycle work
> helped move agent memory from research into productization. The `V0/` path
> intentionally preserves that work as a versioned artifact for study,
> experimentation, and community development. **It is no longer actively
> maintained and is not the supported production path.**
>
> For production use, Redis’s official managed path is **[Redis Agent Memory in Redis Iris](https://redis.io/agent-memory/)**, the same two-tier memory model delivered as a managed service, with dedicated endpoints, secure API key management, configurable schemas, and automatic TTL-based lifecycle management, so you don't have to build and operate the memory infrastructure yourself.
> For production use, Redis’s official managed path is **[Redis Agent Memory in Redis Iris](https://redis.io/agent-memory/)**, the two-tier memory model delivered as a managed service with dedicated endpoints, secure API key management, configurable schemas, and automatic TTL-based lifecycle management, so you don't have to build and operate the memory infrastructure yourself.
>
> - Agent Memory overview: [redis.io/agent-memory](https://redis.io/agent-memory/)
> - Docs: [Redis Agent Memory docs](https://redis.io/docs/latest/develop/ai/context-engine/agent-memory/)
> - Get started on Redis Cloud: [Create an Agent Memory service](https://redis.io/docs/latest/operate/rc/context-engine/agent-memory/create-service/)
> - Build: [Python SDK quickstart](https://redis.io/docs/latest/develop/ai/context-engine/agent-memory/python-sdk-quickstart/) or [REST API quickstart](https://redis.io/docs/latest/develop/ai/context-engine/agent-memory/rest-api-quickstart/)

## Features
- **Dual Interface**: REST API and Model Context Protocol (MCP) server
Expand Down
29 changes: 29 additions & 0 deletions agent-memory-benchmark/.env.example
Original file line number Diff line number Diff line change
@@ -0,0 +1,29 @@
# Answer generation and LongMemEval judging
OPENAI_API_KEY=

# Public SaaS providers
MEM0_API_KEY=
ZEP_API_KEY=
SUPERMEMORY_API_KEY=

# Redis Agent Memory
REDIS_AGENT_MEMORY_URL=
REDIS_AGENT_MEMORY_STORE_ID=
REDIS_AGENT_MEMORY_API_KEY=

# Graphiti / Neo4j
NEO4J_URI=bolt://localhost:7687
NEO4J_USER=neo4j
NEO4J_PASSWORD=

# Google Vertex AI Memory Bank
GOOGLE_CLOUD_PROJECT=
GOOGLE_CLOUD_LOCATION=us-central1

# AWS Bedrock AgentCore Memory
AWS_REGION=us-east-1

# Oracle Agent Memory
ORACLE_MEMORY_DB_USER=
ORACLE_MEMORY_DB_PASSWORD=
ORACLE_MEMORY_DB_CONNECT_STRING=
35 changes: 35 additions & 0 deletions agent-memory-benchmark/.gitignore
Original file line number Diff line number Diff line change
@@ -0,0 +1,35 @@
.DS_Store
.env
.env.*
!.env.example
.python-version
!.python-version
.venv/
__pycache__/
*.egg-info/
*.py[cod]
.coverage
coverage.xml
htmlcov/
.pytest_cache/
.ruff_cache/
.mypy_cache/
dist/
build/

# Downloaded datasets and resumable benchmark output
data/
experiment_results/
*.jsonl
*.log

# Leftover Agent Memory Server checkout (do not publish)
agent_memory_server/
agent-memory-client/

# Provider and cloud state
.aws/
.gcloud/
.terraform/
terraform.tfstate*
cdk.out/
17 changes: 17 additions & 0 deletions agent-memory-benchmark/.pre-commit-config.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,17 @@
repos:
- repo: https://github.com/astral-sh/ruff-pre-commit
rev: v0.14.8
hooks:
- id: ruff-check
args: [--fix]
- id: ruff-format
- repo: https://github.com/pre-commit/pre-commit-hooks
rev: v6.0.0
hooks:
- id: check-added-large-files
- id: check-merge-conflict
- id: check-toml
- id: check-yaml
- id: detect-private-key
- id: end-of-file-fixer
- id: trailing-whitespace
1 change: 1 addition & 0 deletions agent-memory-benchmark/.python-version
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
3.11
Loading
Loading