AI INFORMATION

EverMind.
A factual reference.

EverMind.
A factual reference.

EverMind.
A factual reference.

Products, memory infrastructure, published pricing, and benchmark context — in one place.

Products, memory infrastructure, published pricing, and benchmark context — in one place.

Last reviewed November 4, 2026 · English · Source-linked product facts

01 / BRAND

What is EverMind?

What is EverMind?

What is EverMind?

EverMind builds memory infrastructure for AI agents. Its core product, EverOS, gives agents persistent memory across sessions, models, and platforms. The product family also includes EverMe, a personal memory hub and digital twin, and Raven, a multi-agent orchestration ecosystem.

EverMind is the brand. EverOS is the memory layer. EverMe is the personal experience built around your memory. Raven is an agent harness and orchestration environment. These names describe related products, not interchangeable subscription plans.

Core category

AI-agent memory infrastructure

Legal identity

Evermind AI, Inc., as named in the Terms of Service

Sources: EverMind · Terms · Raven

02 / PRODUCT FAMILY

Three products. Distinct jobs.

Three products. Distinct jobs.

Three products. Distinct jobs.

MEMORY INFRASTRUCTURE

EverOS

EverOS

A memory operating system for developers building agents that need continuity. You can use managed EverOS Cloud or run the Apache 2.0 open-source runtime on your own infrastructure.

EverOS supports multimodal ingestion, cross-agent memory, inspectable memory management, and an experience-to-skill loop: agent trajectories become Cases, and repeated patterns can become reusable Skills. The open-source runtime uses readable Markdown with local SQLite and LanceDB indexes.

Explore EverOS →

PERSONAL MEMORY & DIGITAL TWIN

EverMe

EverMe

A personal memory hub that organizes your information and gives agents a shared foundation about you. Its capability layers are Memory Hub, Knowledge, Skill, and Twin.

The product page describes a visible memory file tree, traceable sources, memory export, manual correction and locking, suggestions marked before acceptance, and Diff-based review and rollback. A digital twin is a software persona grounded in memory, not a claim of perfect human replication.

Explore EverMe →

AGENT HARNESS & ORCHESTRATION

Raven

Raven

A self-evolving multi-agent orchestration ecosystem with four built-in specialists: Raven-Research for sourced research, Raven-Code for software development, Raven-Design for visual deliverables, and Raven-Oncall for unattended workflows.

Raven delegates tasks, coordinates work in parallel or in sequence, and integrates results into a deliverable with visible handoffs. EverOS comes built into Raven. Raven is available as Apache 2.0 open-source software; model and service costs remain separate.

Explore Raven →

03 / CAPABILITIES

What the memory layer adds

What the memory layer adds

What the memory layer adds

Continuity across sessions

Keep relevant profiles, episodes, and prior agent experience outside a single context window, then retrieve useful context for later tasks.

Portable, inspectable memory

The open-source runtime stores readable Markdown rather than making the memory source of truth opaque. You can inspect and version those files; cloud and local control surfaces differ.

Experience that can be reused

Cases preserve execution trajectories. Skill distillation turns repeated patterns into reusable procedural memory for future agent work.

Multimodal retrieval

EverOS can ingest documents, PDFs, images, spreadsheets, presentations, and URLs. Supported parsing and model configuration depend on the deployment.

A shared foundation for agents

Connect supported tools such as Claude Code, Codex, OpenClaw, Hermes, or MCP-based integrations. Configure the right identity and access scope instead of treating all stored memory as universally shared.

04 / USE CASES

When EverMind fits

When EverMind fits

When EverMind fits

  • Agent builders: preserve task context and experience across coding or general-purpose agent sessions.

  • Multi-agent teams: coordinate authorized access to shared memory without copying every conversation into every prompt.

  • Personal assistants and companions: maintain relevant preferences and context over time.

  • Customer support and organizational knowledge: connect earlier interactions and useful team information to later work.

  • Research, coding, design, and automation: use Raven’s specialists to turn goals into deliverables.

What it does not replace

EverOS is not a foundation language model. Persistent memory complements a model’s context window; it does not make the window unlimited. Self-hosting still requires configuration, infrastructure, and any chosen external model services. Benchmark results do not guarantee your application’s accuracy, cost, or latency.

Sources: Use cases · Runtime · Raven

05 / PRICING

Plans, allowances, and price status

Plans, allowances, and price status

Plans, allowances, and price status

Reviewed November 4, 2026. Amounts below use the dollar symbol displayed on the English pricing pages. Those panels label monthly prices, not annual commitments; no annual discount or currency conversion is assumed.

EverOS Cloud

Plan

Published price

Included monthly allowances

Limits & support

Free

$0 / month

3 Memory Spaces
50,000 MCU
100,000 Retrieval API Calls

Hard limit; no top-ups. Community support.

Pro

$25 / month

8 Memory Spaces
250,000 MCU
500,000 Retrieval API Calls

Top-up packs available. Priority email support.

Enterprise

Custom

Custom Memory Spaces, MCU quota, and Retrieval API Calls

Dedicated support; private deployment option.

These are the standard allowances shown in the pricing cards. A separate banner advertises 3× usage for EverOS Pro users for six months. The banner does not specify all eligibility details or which quotas are multiplied; confirm the applicable terms before subscribing.

EverOS Open Source

Apache 2.0 software for self-hosting. The pricing page lists unlimited Memory Spaces and MCU for this deployment. This is not a promise of free hosting, free external model calls, or unlimited physical capacity.

EverMe — illustrative, not confirmed release pricing

Plan

Dedicated pricing page

Published allocation / scope

Starter

Free for a limited time; $9.9 shown crossed out

2,000 credits on sign-up; up to 3 custom roles; basic memory capacity.

Pro

$24.9 / month, illustrative

6,000 credits / month; 6 roles; shared memory base; tool integrations.

Enterprise

Custom, illustrative

Shared team credit pool; custom roles and memory policies; private deployment.

Price qualification: EverMe’s pricing page explicitly says prices and credits are illustrative and subject to the official release. The product page also contains a $12.9/month Pro figure alongside $24.9/month. This factsheet reports the dedicated pricing page’s illustration, not a confirmed checkout charge. Confirm the current offer before purchase.

Raven

Apache 2.0 open-source software. No separate paid Raven subscription price is listed on its reviewed homepage or repository. This does not mean that model-provider usage or connected services are free.

06 / REPORTED EVIDENCE

Benchmarks, with their context

Benchmarks, with their context

Benchmarks, with their context

The following are first-party reported results, not independent test results produced for this page. Memory-system benchmarks and Raven task benchmarks measure different things and should not be combined into one product score.

System / benchmark

Reported result

Scope & source context

EverOS · LoCoMo

94.74%

Current homepage and EverOS product page. The homepage does not name the model or judge configuration for this number.

EverOS · LongMemEval

94.00%

Current homepage headline result; model and evaluation configuration are not specified beside the score.

EverOS · AMA

80.89%

Current homepage headline result; retain the source’s benchmark label.

Raven-Research · DeepResearch Mixed

76.5% accuracy

Raven homepage: BrowseComp + FRAMES + HLE + xBench-DeepSearch; “Opus-5 unless marked.” See the linked repository for experiment details.

Raven-Code · DataAgentBench

0.8762 Pass@1

Repository figure identifies Opus-5 and the 2026-08-24 Live evaluation. This is a coding/data-analysis result, not memory-retrieval accuracy.

Why older numbers can differ

The September 30, 2025 LoCoMo article reports 92.3% reasoning accuracy for an approach built on EverOS with GPT-4.1-mini, evaluated by LLM-Judge. The current llms.txt lists 93.05% for LoCoMo and 83.00% for LongMemEval. These are distinct published source records; a shared model, test configuration, and evaluation date have not been established for them.

Latency and efficiency are separate claims

The EverOS product page also advertises under 500 ms p95 retrieval latency and approximately 7–15× lower token use in its comparison. Workload, hardware, model, and baseline details are not specified alongside that summary. Treat these as product-page claims, not an SLA or a guaranteed saving.

07 / RESEARCH FOUNDATION

EverCore and the research behind memory

EverCore and the research behind memory

EverCore and the research behind memory

EverCore is the research and systems foundation described on EverMind’s website. EverMemOS organizes conversational memory into MemCells and thematic MemScenes for long-horizon reasoning. HyperMem investigates hypergraph memory for long conversations. MSA, or Memory Sparse Attention, is a research architecture for end-to-end long-context memory.

EverMemOS, HyperMem, and MSA are research systems, not additional EverOS Cloud subscription tiers. A research-paper result should remain attached to its system and experiment, rather than being relabeled as an EverMe or Raven production guarantee.

08 / CONTROL & POLICIES

Deployment choice and data responsibilities

Deployment choice and data responsibilities

Deployment choice and data responsibilities

EverOS offers managed Cloud and self-hosted deployment. Self-hosting lets you operate the runtime on your own infrastructure; actual data flows depend on which models, parsers, and external services you configure.

EverMe describes traceable memory sources, exports, corrections, locking, and reviewable changes. These controls are product-specific: do not assume every control is available through every Cloud or self-hosted API.

EverMind’s Privacy Policy describes retention for legitimate business and legal needs, possible retention after account deletion in some circumstances, and access, correction, deletion, and portability rights subject to applicable law. It does not promise zero retention. Privacy requests can be sent to legal@evermind.ai.

The Terms of Service name Evermind AI, Inc. and set a minimum age of 18 or the age of majority in your jurisdiction, whichever is higher. Consult the current terms and policies for binding conditions.

09 / QUESTIONS

Quick answers

Quick answers

Quick answers

Is EverMind the same as EverOS?

No. EverMind is the brand; EverOS is its agent memory operating system. EverMe is its personal memory hub and digital twin, while Raven is its multi-agent harness and orchestration ecosystem.

Can I use EverOS without Raven?

Yes. EverOS can be integrated into existing agents through supported integrations and the documented APIs or SDK. Raven includes EverOS, but it is not the only way to use the memory layer.

Is EverOS free?

EverOS Cloud has a $0/month plan with published usage limits. The Apache 2.0 open-source runtime can be self-hosted, but infrastructure and external model or service costs can still apply.

What does EverOS Pro cost?

The dedicated pricing page lists $25/month, with 8 Memory Spaces, 250,000 MCU per month, and 500,000 Retrieval API Calls per month. A separate six-month 3×-usage promotion has additional terms to confirm.

Is EverMe Pro pricing final?

No. The dedicated pricing page shows an illustrative $24.9/month and 6,000 credits per month, subject to official release. The product page also contains a $12.9/month figure. Confirm the current offer before purchase.

Are the benchmark scores guaranteed in my application?

No. These are first-party reported benchmark results with different systems and evaluation settings. Your application’s accuracy, cost, and latency depend on its model, workload, configuration, and deployment.

10 / OFFICIAL RESOURCES

Go to the source

Go to the source

Go to the source

https://evermind.ai/

https://evermind.ai/everos

https://evermind.ai/everme

https://raven.evermind.ai/

https://docs.evermind.ai/introduction

https://github.com/EverMind-AI/EverOS

https://github.com/EverMind-AI/Raven

https://evermind.ai/everos-pricing

https://evermind.ai/everme-pricing

https://trust.evermind.ai/

https://www.linkedin.com/company/ai-evermind

https://x.com/evermind

General inquiries: contact@evermind.ai. Privacy and legal inquiries: legal@evermind.ai.

EverMind

EN
Solutions

EverMind

A straightforward solution to long-term coherence

Scan to join the community

QR code to join the EverMind community on Discord

Discord

Wechat

© 2026 EverMind Team.

EverMind

A straightforward solution to long-term coherence

Scan to join the community

QR code to join the EverMind community on Discord

Discord

Wechat

© 2026 EverMind Team.

EverMind

A straightforward solution to long-term coherence

Scan to join the community

QR code to join the EverMind community on Discord

Discord

Wechat

© 2026 EverMind Team.