Memory & knowledge for MCP agents
Your AI agent, with a memory that sticks.
Persistent memory and grounded knowledge from your own docs — as native MCP tools your agent calls directly. Drop in any file, URL, or YouTube video; skip building and babysitting a RAG stack. One API, no re-pasting, no token waste.
Start free in 30 seconds — no credit card, no sales call.
Built for every model
OpenAI
Claude
Gemini
Meta
Mistral
Perplexity
DeepSeek
Cohere
Grok
QwenOpenAIClaudeGeminiMetaMistralPerplexityDeepSeekCohereGrokQwenOpenAIClaudeGeminiMetaMistralPerplexityDeepSeekCohereGrokQwenOpenAIClaudeGeminiMetaMistralPerplexityDeepSeekCohereGrokQwenOpenAIClaudeGeminiMetaMistralPerplexityDeepSeekCohereGrokQwenOpenAIClaudeGeminiMetaMistralPerplexityDeepSeekCohereGrokQwenThe problem
Your AI forgets everything.
Great model, zero memory. Sound familiar?
Every session starts from zero
Your agent re-asks what it learned yesterday.
Your documents are trapped in PDFs
Formats your model can't read are knowledge it doesn't have.
Your RAG stack is 5 services and a prayer
Vector DB, parser, embedder, reranker — all yours to babysit.
Context is scattered everywhere
Docs in one tool, chats in another, memory nowhere.
The fix
One API that gives your AI a memory.
Nine problems your stack has. One platform that solves them.
Memory your agent calls itself
“Goldfish memory kills good agents.”
remember, recall, and search — as MCP tools your agent uses mid-conversation. It keeps users, preferences, and decisions across every session.
- remember / recall / search over REST or MCP
- Auto-created Memory knowledge base
- Near-duplicate dedup · every memory versioned
Drop in any doc, get it grounded
“If your AI can't read it, it can't use it.”
Any file or URL in, clean Markdown out — chunked, embedded, and searchable. The convert-to-ground flow, without a stack to build.
- PDF, Word, Excel, PowerPoint, CSV, HTML, images (OCR), YouTube
- Batch up to 25 · URL fetch · same-domain crawl
- Scheduled URL refresh — keep sources current automatically
New
Chat with YouTube videos
“Hours of video your AI can't read.”
Paste any YouTube link and the transcript becomes a document your AI can search, cite, and chat with.
- Lectures, talks, podcasts, tutorials — any captioned video
- Transcript lands in a knowledge base like any doc
- Ask questions, get answers with citations
Semantic search
“Keyword search misses how people actually ask.”
Find by meaning, not just matching words — hybrid retrieval that combines vectors and full-text, then reranks.
- Vector embeddings catch paraphrases and intent
- Full-text catches exact terms, codes, and names
- Reranking puts the best passage first
Knowledge bases it actually cites
“Scattered docs mean ungrounded answers.”
Group documents into searchable KBs your AI answers from — with sources, not guesses.
- Grounded chat with cited answers
- Feedback corrections that stick
- Visual graph of connected docs · @mention to scope
MCP-native, wired in one command
“Integrations shouldn't take a sprint.”
The same tools work in Claude, Cursor, and any MCP agent — one install, any model.
- /plugin marketplace add l33tcy/memniva-claude
- Agent self-setup at memniva.com/llm-setup
- Same tools over REST
Up to 90% fewer tokens
“Every token you send is cost and latency.”
Your agent pulls only the passages it needs — never a window-burning document dump. Measured up to 90% fewer tokens on real workloads.
- Targeted semantic search instead of whole-KB dumps
- Compact cited results — no JSON bloat or unused metadata
- Lower cost, lower latency, sharper answers
Retrieval built in-house
“Most RAG platforms are thin API wrappers.”
A hybrid retrieval stack built end to end — not a wrapper around someone else's vector DB.
- Hybrid vector + keyword, fused with Reciprocal Rank Fusion
- In-house relevance scoring and reranking
- Fast enough to sit in the request path
Private by default
“Your data isn't someone's training set.”
Your documents stay yours — encrypted, isolated, and never training data.
- Stored files encrypted at rest
- API keys hashed · spaces isolate projects
- Delete means gone — no shadow copies
New
Clean Markdown from any URL.
Turn any URL into clean Markdown or structured data for AI agents — the article, not the nav bars, ads, and footers around it.
- Works on the real web: JS-heavy, gated, and region-specific sites
- Main-content extraction — clean Markdown, not raw HTML
- Works exactly like files: same output quality, everywhere you use us
https://example.com/blog/building-with-ai .md
# Building with AI Clean, structured Markdown — ready to feed straight into your **LLM, RAG, or agent**. ## Why it matters - No nav, ads, or cookie banners - Tables preserved as Markdown - Works on JS-heavy pages
💡 Use cases
Built for real AI workflows
RAG pipelines
Clean, chunk-ready Markdown for vector stores and retrieval.
AI agents
Feed agents structured context instead of raw PDFs.
Invoice & form extraction
Pull typed JSON fields with your own schema.
Fine-tuning datasets
Turn document libraries into clean training data.
Knowledge bases
Make internal docs searchable and answerable.
Batch processing
Convert at scale via the API with per-key rate limits.
💬 Loved by builders
Developers ship faster with Memniva
“Pointed the MCP server at Claude and my agent had persistent memory in under five minutes. No SDK, no glue code.”
Sarah K. — ML Engineer
“KB chat cites the exact paragraph it pulled the answer from. Our support bot went from plausible to trustworthy.”
Miguel R. — Founder, devtools startup
“Fed it a 200-page PDF full of tables and got Markdown clean enough to ship straight into our RAG pipeline.”
Priya N. — Data Engineer
“Point it at a URL and it comes back as clean Markdown, not a wall of HTML. I deleted a thousand lines of scraping code.”
Tom H. — Backend Developer
“Spaces keep every client project isolated — each agent only sees its own knowledge bases. Exactly how it should work.”
Aisha B. — AI Consultant
“We replaced a homegrown vector-store setup with knowledge bases. Retrieval got faster and I decommissioned three services.”
Daniel W. — Staff Engineer
“Scanned invoices → OCR → structured JSON with our own schema. It just works, batch after batch.”
Elena V. — Automation Lead
“Best conversion quality I've tested — headings, tables, and code blocks all survive the trip to Markdown.”
Jonas M. — Technical Writer
“Our agents call one API for memory, conversion, and search. One key, one bill, a lot less yak-shaving.”
Ryan C. — Indie Hacker
“Cited answers made compliance review painless — every claim in KB chat links back to the source document.”
Fatima A. — Solutions Architect
“Pointed the MCP server at Claude and my agent had persistent memory in under five minutes. No SDK, no glue code.”
Sarah K. — ML Engineer
“KB chat cites the exact paragraph it pulled the answer from. Our support bot went from plausible to trustworthy.”
Miguel R. — Founder, devtools startup
“Fed it a 200-page PDF full of tables and got Markdown clean enough to ship straight into our RAG pipeline.”
Priya N. — Data Engineer
“Point it at a URL and it comes back as clean Markdown, not a wall of HTML. I deleted a thousand lines of scraping code.”
Tom H. — Backend Developer
“Spaces keep every client project isolated — each agent only sees its own knowledge bases. Exactly how it should work.”
Aisha B. — AI Consultant
“We replaced a homegrown vector-store setup with knowledge bases. Retrieval got faster and I decommissioned three services.”
Daniel W. — Staff Engineer
“Scanned invoices → OCR → structured JSON with our own schema. It just works, batch after batch.”
Elena V. — Automation Lead
“Best conversion quality I've tested — headings, tables, and code blocks all survive the trip to Markdown.”
Jonas M. — Technical Writer
“Our agents call one API for memory, conversion, and search. One key, one bill, a lot less yak-shaving.”
Ryan C. — Indie Hacker
“Cited answers made compliance review painless — every claim in KB chat links back to the source document.”
Fatima A. — Solutions Architect
How it works
From file to AI in three steps
01
Upload
Drop a document — or call the API.
02
Convert
Get clean Markdown or schema-shaped JSON.
03
Ship
Feed it straight into your LLM, RAG, or agent.
Security & trust
Built for production. Private by default.
Enterprise-grade handling of your data — encrypted, isolated, and private by default.
Private by default
Your data is yours — never a training set, never resold.
Isolated infrastructure
Runs on dedicated, isolated infrastructure — no shared tenancy.
Encrypted at rest
Stored files and sensitive credentials are encrypted at rest.
Scoped API keys
Keys are hashed, per-key rate-limited, and revocable anytime.
Project isolation
Spaces keep each project's knowledge separate — no cross-over.
Delete means gone
Remove data and it's gone for good — no shadow copies.
Pricing
More for less
Start free. Upgrade when you grow. Cancel anytime.
Freemium 🎁
$0
- 💬 Chat with your documents — 20 messages a month
- 📄 Turn 10 files or web pages into clean text
- ▶️ Turn 5 videos into searchable notes
- 🧠 1 knowledge base to keep it all in
Start free — in seconds
No credit card required
Most Popular
Pro ⚡
Invite only
- 💬 Chat with your documents — 1,500 messages a month
- 📄 Convert 2,000 files or web pages
- ▶️ Turn 150 videos into searchable notes
- 📚 25 knowledge bases · 25 GB of storage
Request access
Invite-only — we'll be in touch
Business 🚀
Invite only
- ✅ Everything in Pro, with room to grow
- 💬 Chat with your documents — 10,000 messages a month
- 📄 Convert 20,000 files or web pages
- ▶️ Turn 1,000 videos into searchable notes
Request access
Invite-only — we'll be in touch
Get started
Install in under a minute
Claude Code plugin, any MCP client, or plain REST — same tools, same key.
Claude Code — plugin marketplace
/plugin marketplace add l33tcy/memniva-claude /plugin install memniva@memniva export MEMNIVA_API_KEY=kfa_...
Any MCP client
{
"mcpServers": {
"memniva": {
"type": "http",
"url": "https://memniva.com/api/mcp",
"headers": { "x-api-key": "kfa_..." }
}
}
}REST API
curl -X POST https://memniva.com/api/v1/memory \
-H "x-api-key: kfa_..." \
-H "content-type: application/json" \
-d '{"content":"User prefers concise answers."}'Agents can self-configure from memniva.com/llm-setup · full API reference
Frequently asked
Give your AI a memory.
Your agent forgets nothing from here on. Upgrade when you scale.
Start FreeStart free in 30 seconds — no credit card, no sales call.