Vektori
Open-source graph memory engine that gives AI agents sentence-level recall across sessions
Vektori solves the part of agent memory that flat vector stores quietly fail at: multi-hop, time-aware recall with citations. L0/L1/L2 query depths are genuinely useful when you need to budget latency per call. The price of entry is real — you pick and operate a database, and Python is assumed. If you want memory hosted for you, this isn't it.
Verified 1d ago · liveness 59/100 · cite: rightaichoice.com/tools/vektori
- AI agent developers who need persistent, contextual memory across sessions
- Conversational AI engineers building support bots with long customer histories
- Multi-agent architects wanting a shared, isolated memory layer
- AI tutors that must track a student's learning history over weeks
- Teams that need a fully managed, cloud-hosted memory service
- Enterprise buyers who require vendor-backed compliance certifications or SLAs
- Teams without Python skills or anyone to own database infrastructure
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 Vektori if you want a hosted memory service with an SLA and no infrastructure to run, or if flat key-value memory already covers your agent's recall needs.
Self-hosting is free software but not free infrastructure—you pay for the server, database, and vector store you run Vektori on.
Vektori itself is $0/mo under Apache 2.0—the real budget line is the infrastructure and model tokens you supply. That suits startups and research teams comfortable running their own Postgres or vector database, and it undercuts hosted memory services that charge per stored memory or per API call. Larger enterprises that need vendor-managed uptime, certifications, or support contracts will spend more elsewhere and get less operational burden in return.
In short
Vektori — Open-source graph memory engine that gives AI agents sentence-level recall across sessions. Best for AI agent developers who need persistent, contextual memory across sessions, Conversational AI engineers building support bots with long customer histories, Multi-agent architects wanting a shared, isolated memory layer. Free to use.
What people actually say about Vektori — 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.
2 mentions across 2 sources (GitHub, Lemmy) · researched Jul 3, 2026.
Average across the 2 sources that answered — each source counts once, not each post.
- +Innovative three-layer sentence graph for rich memory.
- +High retrieval accuracy (~95%) on standard benchmarks.
- +Open-source Apache 2.0 license with free use.
- +Flexible storage backends from SQLite to Neo4j.
- +Supports multiple LLM providers including local models.
- −Very early stage with limited community validation.
- −Only two relevant community posts available for analysis.
- −25 open issues may indicate stability concerns.
- −No real-world production case studies shared.
- −Graph database setup adds complexity for production.
- • Self-hosting costs for graph databases (e.g., Neo4j cloud subscription)
- • API costs from LLM providers when using cloud models
Viability Score
How well maintained and how widely used is Vektori? 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
- Sentence-level text splitting that preserves semantic boundaries
- Three-layer memory graph of Facts, Episodes, and Sentences
- Personalized PageRank retrieval ranked with temporal decay
- Three query depths: L0 facts, L1 facts plus episodes, L2 full trajectory
- Four-tier hierarchy: Sentences, Facts, Insights, and Summaries
- Grounded retrieval that cites the source conversation as evidence
- Pattern discovery across multiple sessions
- Session-level and user-level memory isolation
- SQLite default backend for zero-config local development
- Production storage on PostgreSQL/pgvector, Neo4j, Qdrant, or Milvus
- In-memory backend for CI and testing
- Python-first API with quickstart examples
- Benchmarking suite for LoCoMo and LongMemEval-S
- Model-agnostic LLM and embedding access via OpenAI, Azure, Anthropic, NVIDIA, or LiteLLM
- Open-source Apache 2.0 license for self-hosting
About Vektori
Vektori targets a familiar failure in AI agents: the bot remembers a fact about a user but loses the story around it. Its answer is a self-hosted memory layer under Apache 2.0 that splits text at sentence boundaries and stores what it learns in a three-layer graph of Facts, Episodes, and Sentences. An agent can then answer what happened, when, and why rather than retrieving a near-match chunk and hoping. Retrieval runs on Personalized PageRank. Seed sentences come from vector similarity, get expanded through the graph, then ranked with temporal decay. Query depth is a per-call choice: L0 returns facts only, L1 adds episodes, L2 returns the full trajectory with cited source conversations. A fourth tier of Sentences, Facts, Insights, and Summaries sits above that for pattern discovery across sessions. The infrastructure is deliberately your problem, not the vendor's. SQLite is the zero-config default, an in-memory backend covers CI, and production leans on PostgreSQL/pgvector, Neo4j, Qdrant, or Milvus. Models route through OpenAI, Azure OpenAI, Anthropic, NVIDIA, or local models via LiteLLM. Session-level and user-level isolation mean multi-agent setups can share one layer without bleeding context, and a benchmarking suite ships for LoCoMo and LongMemEval-S. Mem0 and LangMem get an agent remembering in an afternoon; Vektori gets it recalling relationships. That trade favors Python teams with someone to own a database, and it is the wrong tool for anyone who wants memory as a managed endpoint.
Behind the Verdict
Most agent memory tools optimize for the first session. Vektori optimizes for the tenth, which is where support bots and tutors actually break — the customer's history is a sequence of events, not a bag of facts. We'd reach for it when recall quality is the bottleneck. A tutoring agent that needs to know a student struggled with fractions in week two and again in week five is exactly the L2 use case. So is a multi-agent system where agents share memory but must not leak one user's context into another's. The per-call query-depth control is the detail I like most: cheap fact lookups on the hot path, full trajectory only when the conversation demands it. Pass on Vektori if memory is a feature you want solved, not owned. Choosing between pgvector, Neo4j, Qdrant, and Milvus is a decision with operational consequences, and nobody on the team wants to be the person paged when the graph store fills up. TypeScript-first shops should also look elsewhere — the API is Python. Against Mem0, the honest split is simplicity versus structure. Mem0-style memory gets you running fast and stays out of the way; Vektori buys you relational, temporally-decayed recall and makes you pay in setup and maintenance. The LoCoMo and LongMemEval-S benchmark suite is the tiebreaker worth using — run it on your own data before you commit to either. One caveat in practice: graph memory quality tracks extraction quality. Sentence splitting and fact/episode construction depend on the models you wire in, so budget time for tuning that pipeline before you blame retrieval.
Researching Vektori? 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 Vektori actually fits — and what changes day-one when you adopt it.
Clones the repo, runs the Python quickstart with the SQLite backend and the in-memory option for tests, and wires Vektori behind a personal assistant so it remembers the user's preferences and past decisions across sessions.
Outcome: Working cross-session recall locally in an afternoon with zero database setup, upgradeable later to a production backend.
Deploys Vektori on PostgreSQL/pgvector, ingests months of ticket transcripts, and queries at L1 (facts + episodes) so the bot knows the history of each customer's issues before replying.
Outcome: Support agents stop re-asking questions customers already answered, and every recalled fact cites the originating conversation.
Gives each agent its own isolated memory namespace and shares an Episode layer so a planner agent can see what a researcher agent learned, using L2 trajectory queries to audit how a decision was reached.
Outcome: Agents coordinate on shared context without leaking one user's memory into another's.
Use Cases
- Persist user preferences across conversations in a personal AI assistant
- Retrieve customer history and past issues in a support bot without asking redundant questions
- Adapt tutoring explanations based on what a student has struggled with previously
- Share memory among multiple AI agents in a collaborative system
- Analyze behavioral shifts in user communication style over time
- Replay full conversation trajectories to understand how decisions evolved
Models Under the Hood
as of 2026-09-29
Limitations
- Vektori is self-hosted; there is no managed cloud tier, so you handle deployment, scaling, backups, and upgrades.
- Retrieval quality depends heavily on the LLM and embedding models you choose and on your hardware.
- Documentation is still evolving, and framework integration is limited to the Python SDK and an MCP server—there is no drop-in LangChain or CrewAI adapter.
- You must pick and operate a storage backend yourself (SQLite for local work, PostgreSQL/pgvector, Neo4j, Qdrant, or Milvus for production).
- No compliance certifications or enterprise SLA are published.
as of 2026-09-29
Verification history
We have re-verified Vektori 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-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
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.
Plans compared
For each published Vektori tier: who it actually fits, and what it adds vs. the previous tier. Cross-reference the cost calculator above for projected annual outlay.
Open Source (Self-Hosted)
$0/mo
Ideal for
Python-skilled agent developers, research teams, and startups who can run their own database and want memory data to stay on their own infrastructure.
What this tier adds
Starting tier—the only tier. Full three-layer graph engine under Apache 2.0 at $0/mo; you supply models, servers, and database operations.
Where the pricing makes sense
The company stage and team size where Vektori's pricing actually pencils out — and where peers do it cheaper.
Vektori itself is $0/mo under Apache 2.0—the real budget line is the infrastructure and model tokens you supply. That suits startups and research teams comfortable running their own Postgres or vector database, and it undercuts hosted memory services that charge per stored memory or per API call. Larger enterprises that need vendor-managed uptime, certifications, or support contracts will spend more elsewhere and get less operational burden in return.
Setup time & first value
How long it actually takes to get something useful out of Vektori — broken out by persona, not the marketing-page minute.
Indie developers: minutes to a working local demo using the Python quickstart with SQLite or the in-memory backend. Engineers wiring it into a real product: a day or two, mostly spent choosing and provisioning a production backend (PostgreSQL/pgvector, Neo4j, Qdrant, or Milvus) and picking your LLM and embedding providers. Multi-agent teams: add time for designing memory isolation boundaries and
Switching to or from Vektori
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From a plain vector store: wrap your existing embeddings as sentence nodes and let Vektori build the Fact and Episode layers over them.
- →From Mem0 or LangMem: re-ingest source conversations through Vektori's sentence splitter, since its graph structure is built at write time rather than derived from flat records.
- →From SQLite to production: keep the same API and swap the backend to PostgreSQL/pgvector, Neo4j, Qdrant, or Milvus.
- ↗To a hosted memory service: export the underlying sentence and fact records from your database backend, since the graph lives in storage you control.
- ↗To a flat vector store: flatten Facts and Episodes back into standalone records, accepting the loss of temporal and relational ranking.
Integrations
Resources & Guides
Tutorials & Learning
YouTube returned 6 videos for “Vektori”, and we withheld 6: 6 could not be judged, because “Vektori” 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 Vektori.
Official links
Tools that pair well with Vektori
Common stack mates teams adopt alongside Vektori, with the specific reason each pairing earns its keep.
Cognee
Cognee is an open-source AI agent memory platform that stores facts in a knowledge graph and returns cited recall across sessions.
Agentmemory
Open-source memory runtime that gives coding agents persistent recall between sessions, stored on your machine.
Powermem
Open-source, self-hosted memory layer that gives AI agents persistent recall via hybrid retrieval.
Featured Head-to-Head Comparisons
Vektori vs Spider Cloud
Choose Spider Cloud if your bottleneck is gathering fresh web data for RAG or AI agents quickly and cheaply; choose Vektori if your bottleneck is remembering conversation history and user preferences over time. They complement each other — Spider Cloud feeds Vektori's memory graph with live data.
Vektori vs Temporal Ai
Choose Temporal AI if you need rock-solid failure recovery for AI agents or microservices orchestration with multi-language support. Choose Vektori if you're building a Python-based conversational AI that requires a long-term, graph-based memory layer to track user context and preferences. Both are open-source, but serve fundamentally different needs.
Vektori vs Presto Voice
If you run a QSR chain looking to automate drive-thru ordering and boost revenue, Presto Voice is a proven, enterprise-ready solution with recent high-profile adoption (e.g., Dairy Queen). For AI developers building agents that need sophisticated, graph-based memory, Vektori is a powerful open-source framework. Choose based on your domain: restaurant operations vs. conversational AI development.
Alternatives to Vektori
View all
Cognee
Cognee is an open-source AI agent memory platform that stores facts in a knowledge graph and returns cited recall across sessions.
Agentmemory
Open-source memory runtime that gives coding agents persistent recall between sessions, stored on your machine.
Frequently Asked Questions
Used Vektori? Help shape our editorial sentiment research.