OttO
OttO is a local-first MCP memory server and knowledge graph that gives Claude Code and Cursor persistent agent memory.
If your agent forgets everything between sessions, OttO is the most direct fix we've seen: a real knowledge graph on your own disk, an MCP endpoint your client already speaks, and a free Personal tier with no node cap. The 55% token cut on search output and the cheap locate tool matter more than they sound — memory calls stop eating your context budget. v1.0.1's Decision nodes finally capture the why, not just the what. Budget setup time: you're wiring a daemon and MCP JSON yourself.
Verified 6h ago · liveness 81/100 · cite: rightaichoice.com/tools/otto
- Developers running Claude Code, Cursor, Antigravity or Codex who want agent memory that survives across sessions
- Engineers who want past decisions and the reasons behind them searchable instead of re-derived
- Privacy-first teams self-hosting one shared knowledge graph with RBAC, SSO/SAML and audit logs
- Solo builders who want unlimited knowledge nodes on the free Personal tier and run Ollama locally
- Non-technical users who want a zero-setup notes app with no daemon or MCP client config
- People who need cloud sync between devices without installing a background service
- Anyone expecting a traditional CMS or WYSIWYG wiki editor
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip OttO if you want a note app that syncs across devices with zero setup — OttO needs a background daemon installed as a system service, an MCP client config, and an embedding-provider choice before it does anything.
Personal is free forever with unlimited knowledge nodes, so single developers pay nothing beyond their own Ollama or OpenAI/Groq API keys. Self-hosted team plans bill annually and scale by seat count — Team covers up to 10 seats at $2,000/yr, Business up to 25 at $4,500/yr, Business+ up to 50 at $8,000/yr, and Enterprise starts at $15,000/yr for 100+ seats. The team tiers include the graph, RBAC, SSO/SAML, audit logs and self-hosted or custom deployment. Compare against per-seat AI memory and
In short
OttO — OttO is a local-first MCP memory server and knowledge graph that gives Claude Code and Cursor persistent agent memory. Best for Developers running Claude Code, Cursor, Antigravity or Codex who want agent memory that survives across sessions, Engineers who want past decisions and the reasons behind them searchable instead of re-derived, Privacy-first teams self-hosting one shared knowledge graph with RBAC, SSO/SAML and audit logs. Free to start; paid plans from $2,000/yr.
What's new in OttO
Checked 6 days agoAcross the latest 3 updates: 2 feature updates and 1 launch.
v1.0.1 — Decisions & Attached Files
Adds Decision nodes with save_decision, search_decisions and set_decision_status, record_discussion, opt-in chat auto-capture, PDF/Word/text attachments readable via list_files and read_file, a multi-select Nodes screen and per-tool toggles in Settings.
v0.2.0 — Task Queue & Plans
Adds a desktop task queue with enqueue, get_next_task and complete_task MCP tools plus REST endpoints, Plan nodes, a locate tool at roughly 10x lower cost than spread search, and a compact search format cutting tokens about 55%.
v0.1.0 — Initial Release
First release of OttO Core (local vector knowledge graph with cosine similarity search) and OttO Desktop, with knowledge and inference node types, Ollama and OpenAI embeddings, and the OTTODB02 storage format.
What people actually say about OttO — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
14 mentions across 4 sources (Hacker News, YouTube, Product Hunt, Lemmy), 69 more we could not attribute · researched Sep 22, 2026.
Weighted by the 83 posts each of 4 sources contributed.
- +Local-first architecture keeps notes and graph data private, avoiding cloud RAG exposure
- +Automatic semantic linking turns isolated notes into a connected, self-updating graph
- +MCP server support targets Claude and Cursor directly, matching modern agent workflows
- +Free personal tier includes unlimited knowledge nodes, lowering trial cost to zero
- +BYO API keys across Ollama, OpenAI, and Groq give users real model flexibility
- −No independent community reviews substantiate the reliability or quality claims
- −Keyword collisions with insurance, films, and a GIF camera drown out real discussion
- −No evidence surfaced on MCP latency or correctness with Claude or Cursor
- −Bring-your-own-key model means agent enrichment costs scale with usage, not fixed
- −Enterprise pricing, RBAC details, and SLAs are entirely unspecified in public data
- • LLM provider API costs for agent reasoning are borne by the user via BYO keys
- • Enterprise pricing is undisclosed, so budget planning requires a sales conversation
- • No community data on whether heavy graph enrichment meaningfully increases token spend
Viability Score
How well maintained and how widely used is OttO? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →Key Features
- Local-first MCP memory server at http://localhost:47821/mcp (HTTP transport)
- REST API at http://localhost:47821/api for programmatic read and write
- 24 MCP tools spanning knowledge, semantic and graph-spreading search, inferences, plans, decisions, files and tasks
- Knowledge graph auto-classifies each note and links it to related nodes
- Decision nodes: save_decision, search_decisions, set_decision_status with proposed/accepted/rejected/superseded status
- record_discussion saves conclusions and decisions from a multi-agent or user discussion in one call
- Opt-in chat auto-capture persists decisions and conclusions after substantive chat turns
- Task queue with enqueue, get_next_task and complete_task MCP tools plus Queue REST endpoints
- Plan nodes with save_plan and search_plans
- Inference nodes record solved problems and the reasons behind them
- Attached PDF, Word and text files chunked into search and readable via list_files / read_file
- locate tool for 'where is X' queries, reported ~10x cheaper than spread search
- Compact [ID|role|score] search output, roughly 55% fewer tokens per search
- Built-in agent answers questions over your graph using Ollama, OpenAI or Groq
- Ollama (fully local) or OpenAI embeddings, switchable under Settings → Provider
About OttO
OttO is a local-first MCP memory server and knowledge graph built for AI coding agents. Install OttO Core — a Rust background daemon — and it runs an MCP server on localhost at http://localhost:47821/mcp plus a REST API on the same port. Claude Code, Cursor, Antigravity or any HTTP-transport MCP client then queries your own knowledge graph directly, with no export step and no copy-paste between chats. Nothing is stored on OttO's servers; with Ollama handling embeddings and chat, no text leaves your machine at all. Every note you add becomes a node the system classifies, embeds and links to related ideas on its own — you explore the graph instead of maintaining it. OttO Core exposes 24 MCP tools covering knowledge capture, semantic and graph-spreading search, inferences (solved problems and their reasons), plans, decisions, attached files and a shared task queue. A built-in agent answers plain-language questions over your nodes using Ollama, OpenAI or Groq, and writes conclusions back into the graph. The recent releases pushed memory closer to how engineering work actually happens. v0.2.0 added Plan nodes, a task queue with enqueue, get_next_task and complete_task, a compact [ID|role|score] search output that cuts tokens per search by roughly 55%, and a locate tool reported around 10x cheaper than spread search for "where is X" lookups. v1.0.1 (September 30, 2026) introduced Decision nodes with save_decision, search_decisions and set_decision_status, record_discussion for capturing a multi-agent conversation's conclusion in one call, opt-in chat auto-capture, and PDF, Word and text attachments readable via list_files and read_file. OttO Desktop is a Tauri + React graph explorer for macOS, Windows and Linux, and OttO Mobile connects to the same graph from iOS or Android. Personal use is free forever with unlimited nodes and bring-your-own keys. The closest alternative is a plain vector store behind a single retrieval tool — that returns isolated chunks, whereas
Behind the Verdict
OttO earns its place when a project outlives a single session. Refactors, incident debugging, multi-week research — anything where you keep re-explaining the same context to Claude Code — is exactly the scenario a local graph is built for. We'd reach for it before a cloud memory service because the data stays on disk and the SQLite-free setup with Ollama means nothing leaves the laptop. The honest cost is setup. You install OttO Core, add the HTTP server to .mcp.json (the installer also drops a session hook into ~/.claude/settings.json that reminds Claude Code to use the tools), and pick an embedding provider. That's fine for a developer who already lives in MCP config, annoying for anyone else. Against a plain vector database with one retrieve tool, the difference shows up in graph search: semantic and spread search walk links, so a query about a caching bug can surface the incident note and the decision that followed. Also, decisions and inferences carry status and reasons, which is the part flat memory stores miss entirely. If all you need is chunk retrieval, OttO is more machinery than the job requires. Where it bites: the graph only pays off once it's dense. A fresh install with twelve notes performs like a decent search box, not a second brain. Commit to writing findings into it for a couple of weeks before judging it. Teams should also know the shared, multi-tenant edition is a self-hosted annual purchase, not something you flip on in the Personal app. Provider choice has privacy consequences worth reading twice. Ollama keeps everything local; OpenAI or Groq sends the embedded or answered text to that vendor with your key. Nothing about that is hidden, but pick deliberately. Pass if you want cloud sync with zero background service, a WYSIWYG wiki, or a
Researching OttO? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas OttO actually fits — and what changes day-one when you adopt it.
Installs OttO Core, adds the HTTP server to ~/.claude/mcp.json, and lets the installer's session hook remind Claude Code to use OttO tools. During a refactor, they call save_inference whenever they solve a gnarly problem and search_inferences later instead of retracing steps.
Outcome: Solved problems carry across sessions, and locate answers 'where is X' for about a tenth of the cost of a full spread search.
Uses save_decision to record what was chosen, why, the alternatives and a status, and record_discussion to capture the conclusions of a postmortem call in one call. Later they filter the Nodes screen by decision status and read the decision detail panel.
Outcome: The team can recall the reasoning behind a fix months later, not just the text of it.
Deploys the self-hosted team plan with a shared knowledge graph, RBAC, SSO/SAML and audit logs, and points everyone's Claude Code, Codex, Cursor and Antigravity at the internal MCP endpoint.
Outcome: One graph the whole team writes to, with access control and an audit trail, and no note text leaving their own servers when Ollama handles embeddings and chat.
Use Cases
- Give Claude Code or Cursor persistent memory across sessions so solved problems aren't re-solved
- Record a decision with its alternatives and status, then recall why it was made months later
- Save implementation plans as Plan nodes and search them later by semantic similarity
- Queue up tasks with a topic and description, then work through them in order from the desktop Queue screen
- Ground AI agent answers in your own notes without exporting or copy-pasting files
- Run "where is X" lookups cheaply with the locate tool instead of full spread search
- Attach a PDF or Word spec and let the agent read it by line range mid-session
- Capture conclusions from a multi-agent discussion in one record_discussion call
Models Under the Hood
as of 2026-10-18
Limitations
- OttO Core is explicitly in early access: the docs state APIs and file formats may change between versions.
- It runs as a local daemon (REST API and MCP server on port 47821) and since v0.2.0 also listens on your local network so other devices must present a pairing token, while requests from the machine itself are unaffected.
- Data is stored locally (e.g. %ProgramData%\OttO on Windows), so backups are the user's responsibility.
- Docker is listed as coming soon rather than available, so container-first deployments have to wait.
as of 2026-10-12
Verification history
We have re-verified OttO 10 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-checked, vendor evidence unchanged
- — re-checked, vendor evidence unchanged
- — re-checked, vendor evidence unchanged
Showing the 6 most recent of 10 verification passes.
Free to cite with attribution — this page re-verifies continuously.
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Where the pricing makes sense
The company stage and team size where OttO's pricing actually pencils out — and where peers do it cheaper.
Personal is free forever with unlimited knowledge nodes, so single developers pay nothing beyond their own Ollama or OpenAI/Groq API keys. Self-hosted team plans bill annually and scale by seat count — Team covers up to 10 seats at $2,000/yr, Business up to 25 at $4,500/yr, Business+ up to 50 at $8,000/yr, and Enterprise starts at $15,000/yr for 100+ seats. The team tiers include the graph, RBAC, SSO/SAML, audit logs and self-hosted or custom deployment. Compare against per-seat AI memory and
Setup time & first value
How long it actually takes to get something useful out of OttO — broken out by persona, not the marketing-page minute.
Solo developer on Claude Code: about 10-20 minutes — install the Core package, confirm http://localhost:47821/health responds, add the HTTP server block to .mcp.json, and optionally paste the CLAUDE.md snippet. Cursor or Antigravity: add the same JSON to their config and restart the client. Team self-hosted deployment with SSO/SAML takes longer because it is an annual contract plus a server
Switching to or from OttO
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From a flat notes app or Obsidian vault: add notes through add_knowledge or the Add Knowledge screen and OttO chunks, embeds and links them automatically
- →From a plain vector database: point agents at the MCP server instead of raw similarity calls so results come back linked and ranked with scores
- →From a CLAUDE.md convention file: keep the file as the instruction layer but move the facts themselves into OttO nodes saved with save_inference and save_plan
- →From legacy OttO formats: the OTTODB02 binary format migrates automatically from earlier formats
- →From another MCP memory server: change the .mcp.json server URL to http://localhost:47821/mcp and re-add your saved findings
- ↗To a plain vector database: export your nodes and re-embed them yourself, losing the automatic classification and semantic links
- ↗To a hosted agent memory service: replay saved inferences, plans and decisions through that service's write API; the graph structure will not port
- ↗To a note app or wiki: copy node text out through the REST API, but plan nodes and decision statuses will flatten into plain text
Integrations
Resources & Guides
Tutorials & Learning
YouTube returned 6 videos for “OttO”, and we withheld 6: 6 could not be judged, because “OttO” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about OttO.
Official links
Tools that pair well with OttO
Common stack mates teams adopt alongside OttO, with the specific reason each pairing earns its keep.
Mem0
Mem0 is an AI memory layer that gives agents and apps persistent context across sessions
MemoryLake
MemoryLake is portable AI memory for ChatGPT, Claude, Codex and Gemini — encrypted, exportable, and shared across every agent you run.
Chrome DevTools MCP
Open-source MCP server that gives coding agents live Chrome DevTools for debugging, traces, and browser automation.
Featured Head-to-Head Comparisons
Otto vs Locus Robotics
These tools solve completely different problems: Locus Robotics is for physical warehouse automation to boost picking productivity 2-3x, while OttO is a digital knowledge graph for grounding AI agents in personal notes. Choose Locus if you operate a fulfillment center with fluctuating order volumes; choose OttO if you're a developer needing local persistent memory for AI workflows.
Otto vs Truleo
Choose Truleo if you're a law enforcement agency needing to connect RMS, CAD, jail calls, and body cameras into a single searchable intelligence hub. Choose OttO if you're a developer or tech-savvy user wanting a local-first knowledge graph that grounds AI agents (e.g., Claude via MCP) in your own notes without cloud dependency. They serve entirely different purposes.
Otto vs Presto Voice
Your choice depends on whether you run a QSR drive-thru or build AI agents. Presto Voice is a proven, ROI-driven solution for chains like Dairy Queen, with up to 95% automation and built-in upselling—ideal if you need to scale drive-thru operations and boost revenue. OttO is a free, local knowledge graph for developers who want persistent memory for AI agents; it's perfect if you're building custom agent workflows and value data privacy. They serve completely different markets—pick the one that matches your job.
Alternatives to OttO
View allMem0
Mem0 is an AI memory layer that gives agents and apps persistent context across sessions
MemoryLake
MemoryLake is portable AI memory for ChatGPT, Claude, Codex and Gemini — encrypted, exportable, and shared across every agent you run.
Chrome DevTools MCP
Open-source MCP server that gives coding agents live Chrome DevTools for debugging, traces, and browser automation.
Frequently Asked Questions
Used OttO? Help shape our editorial sentiment research.