Your AI agents'
memory shouldn't
reset every session.
MemQ gives AI agents an intelligence fabric that remembers decisions, retrieves the right context, and shares team knowledge safely across sessions. Stop re-explaining your project. Turn solved problems into reusable working memory.
You are one recall test away from persistent memory · Card required only for the paid 14-day trial · Cancel anytime
{
"mcpServers": {
"memq": {
"type": "http",
"url": "https://mcp.multinex.ai/mcp/v1",
"transport": "streamable-http",
"auth": {
"type": "oauth2",
"discoveryUrl": "https://billing.multinex.ai/.well-known/oauth-authorization-server"
}
}
}
}Your agent auto-discovers 22 hosted MCP tools on connect.
Get API key22
Built-in memory tools
<50ms
Memory recall speed
30s
Setup time
14 days
Free trial
Why MemQ
Agents that remember,
teams that scale.
Remembers work.
Your AI remembers code patterns, decisions, debugging sessions, and architecture choices so you stop re-explaining your project every session.
Chooses the right memory path.
Simple checks stay fast. Recall searches the right namespace. Reflection distills lessons. High-trust operations are guarded before inference runs.
Shares knowledge safely.
When one agent solves a problem, the team can reuse it without leaking private memory. Namespace gates decide who can read or write.
Memory Modes
Private when you need it.
Shared when you want it.
MemQ gives you full control over what stays private, what's shared with your team, and what connects to a global intelligence network.
Private Memory
Your space. Fully isolated.
Every user gets their own private memory space. Your AI remembers your codebase, your preferences, and your decisions — visible only to you. Nobody else can see or search your private memories.
- Isolated per user — nobody else can access
- Your codebase context, patterns, and decisions
- Persists across sessions and projects
Team Memory
Shared intelligence across your organization.
When one person's AI figures something out, every AI on the team knows it. Bug fixes, architecture decisions, onboarding context — it all flows through a shared memory space your entire org can search.
- Shared knowledge plane for your whole team
- One AI learns → every AI benefits
- Org-scoped access control and visibility
Global Intelligence
A knowledge network that gets smarter every day.
Beyond your team, MemQ taps into a continuously-updated global knowledge base. Best practices, security patterns, framework updates, and community-curated solutions — available to every connected AI.
- Curated, continuously-updated knowledge base
- Security patterns, best practices, framework updates
- Your AI starts smarter on day one
How It Works
From connection to compounding
in four steps.
MemQ is not just storage. It is an intelligence fabric for AI agents: write memory, route it safely, recall the right context, and reflect useful patterns into team knowledge.
Write what should not be forgotten.
Decisions become durable memory.
Most memory tools store notes. MemQ stores working knowledge: project decisions, codebase patterns, debugging fixes, team runbooks, preferences, and outcomes. Your agent stops treating yesterday's work like it never happened.
Decision → "Use idempotent retries for billing webhooks" Pattern → "Prefer typed RPC adapters at IO boundaries" Runbook → "If gateway returns 503, retry with jitter"
Route each memory job correctly.
Fast when simple. Guarded when sensitive.
MemQ chooses where a memory belongs, who may access it, and which intelligence path should handle it. Simple validation stays cheap and fast. Retrieval searches the right namespace. Reflection gets deeper reasoning. Billing, VISA, policy, and admin operations are guarded before inference runs.
Fast path → validation, formatting, health checks Recall path → semantic search + namespace context Reflection path → distillation + contradiction repair High-trust path → billing, VISA, policy, admin gates
Recall the right context without waste.
Search working knowledge, not just similar text.
Vector search finds similar text. MemQ combines memory, journal, graph, reflection, and access control so your agent can retrieve the useful decision, procedure, or episode without rereading every file or transcript.
Question → "How did we fix checkout retries?" Recall → matching decision + runbook + prior outcome Result → fewer tokens spent rediscovering old work
Reflect into reusable team intelligence.
Solved once. Reused safely.
Transcript logs remember text. MemQ remembers working knowledge. It reflects on repeated tasks, distills lessons, links related ideas, and keeps private and team memory separated so one solved problem can help the next agent without exposing the wrong namespace.
Agent A writes → "Payment gateway retries on 503" Agent B searches → "503 retry strategy" → Recalls what Agent A learned. Instantly.
Your agent stores decisions, fixes, preferences, and runbooks.
Write, route, recall, reflect. Each loop reduces repeated explanation.
Public Performance Benchmark
MemQ primary@1
0%
Mem0 primary@1
0%
Delta vs Mem0
+0 pts
Leakage-free retrieval
0%
Current public result: MemQ hit primary result rank at 1 in 100% of published retrieval cases, versus 58% for Mem0. Across 36 runs and 12 cases, MemQ also held 100% recall@k, 1.000 MRR, and 100% leakage-free retrieval.
This is an artifact-backed retrieval benchmark with public case files and repeated runs. It is evidence for the hosted memory stack, not a blanket claim about every downstream agent workload.
For users, this means less time paying agents to rediscover prior decisions and more time applying the right context on the first pass.
Avg latency
21 ms
Published MemQ average latency versus Mem0 at 3381 ms.
Tool surface
22
Hosted MCP tools exposed through a single contract.
P95 latency
31 ms
Published MemQ p95 latency with a 3360 ms average advantage over Mem0.
Benchmark corpus
12 cases
3 namespaces across 108 published result artifacts.
Reproducible Benchmark Matrix
Published retrieval results, not marketing estimates.
This table is generated from the same public snapshot used by the proof page and downloads. The Mem0 row refers to the open-source Mem0 adapter in our harness, not Mem0 managed cloud or every possible Mem0 deployment.
| Provider | Runs | Primary@1 | Recall@K | MRR | Leakage-Free | Avg Latency | P95 Latency |
|---|---|---|---|---|---|---|---|
| MemQ MCP | 36 | 100% | 100% | 1.000 | 100% | 21 ms | 31 ms |
| Mem0 OSS adapter | 36 | 58% | 100% | 0.917 | 67% | 3381 ms | 4163 ms |
| Keyword baseline | 36 | 58% | 92% | 0.861 | 67% | 0 ms | 0 ms |
Read the complete reproducible benchmark methodology in the memq-bench repository.
Supported by benchmarks
The pitch, the proof page, and the public repo all point to the same benchmark source.
MemQ is sold with an externally inspectable benchmark trail. Product buyers can move from the promo page to the proof page, the reproducibility hub, and the GitHub repo without losing the artifact path or confusing fixture validation with broader agent-performance claims.
Why it hits harder
Stateless agents lose the thread. MemQ's dual-brain architecture turns prior work into reusable operational memory.
Memory that knows its temperature
HOT recency serves the immediate loop. Semantic recall and exact replay stay behind the same MCP contract instead of leaking storage complexity into the client.
Journal truth with retrievable structure
Soul Journal keeps the append-only truth source. Semantic retrieval, graph recall, and reflection operate as derived layers — not competing authorities.
Commercial control plane already in the path
Hosted access now authenticates against billing-manager entitlements, supports internal-free Multinex users, and emits tool-level usage for product analytics and billing hygiene.
Pricing
Pay for what your agents use.
Four production tiers from read-only exploration to enterprise memory infrastructure.
Free
Personal baseline memory access to try MemQ without checkout.
- Commons read-only access
- 500 reads / month
- 1 concurrent session
- 22 auto-discovered memory tools
- MCP endpoint access
Included at signup. Upgrade when you need private write access.
Contributor
TrialFirst paid solo tier for builders needing writable, persistent memory.
- Everything in Free
- Private namespace (1 included)
- 1,000 writes / month
- 5,000 reads / month
- 1 concurrent session
- Standard gateway inference
- Baseline write validation
Pro
TrialSerious individual tier with stronger retrieval and higher throughput limits.
- Everything in Contributor
- 2 namespaces (commons + private)
- 25,000 writes / month
- 100,000 reads / month
- 3 concurrent sessions
- Priority gateway inference
- Soul Journal (read-only)
- Full L1-L5 write validation
- Anti-poisoning scan + canary
Team
TrialShared organization memory, team-wide retrieval, and admin-managed access.
- Everything in Pro
- Organization shared memory plane
- 150,000 writes / month
- 500,000 reads / month
- 10 concurrent sessions
- Full Soul Journal + export
- Priority gateway inference
- Org-scoped shared namespaces
Managed deployment
Enterprise
Sales-led shared memory, dedicated infrastructure, and custom security controls.
- Shared organization namespaces
- Unlimited namespaces & seats
- Unlimited reads & writes
- Priority+ reserved capacity
- Fail-closed mode by default
- Custom content security rules
Frequently asked questions
Get Started
Give your AI a memory.
Connect in 30 seconds. Start free — upgrade when you're ready.