97.9% session recall on LoCoMo · Up to 90% input token savingsView evals
Open sourceMCP-nativeBuilt on YugabyteDB

The context engine for agentic applications

Shared, self-learning context for every agent, tool and teammate. Better answers, fewer tokens, one governed datapack.

Free plan · no credit card · works with Claude, Cursor, Codex and any MCP client

Claude Code · Cursor · Codex
# Add Meko to any MCP client
claude mcp add --transport http meko \
  https://mcp.mekodata.ai/mcp

# Then, from your agent:
context_search("what did we decide about auth?")
Sign in once; every agent shares the same datapack.Quickstart
Open evals

Context quality on LoCoMo.
A fraction of the tokens.

LoCoMo asks 1,540 questions about ten long conversations, and an LLM judge accepts or rejects each answer (J). The first group uses the standard grader from the Mem0 2025 paper. The second group used each provider’s own grader, so those scores don’t compare directly.

LoCoMo judge accuracy by question type

Provider or runSingle-hop
841 q
Multi-hop
282 q
Open-domain
96 q
Temporal
321 q
Overall J
1,540 q
Standard grader: the LoCoMo paper grader
Meko, run 2gpt-5.6, two-step prompt, paper grader on gpt-4.191.878.769.879.185.4
Meko, run 3gpt-5.3, two-step prompt, paper grader on gpt-4.188.778.071.974.582.7
Lyzr Cognis“GPT-4”, paper grader, overall computed by Meko from its categories85.972.051.077.379.4
Mem0, 2025 papergpt-4o-mini, paper grader72.967.151.255.566.9
Other graders: each provider’s own, not directly comparable
Zepgpt-5.4, own chain-of-thought grader and category counts96.494.079.295.694.7
Meko, baseline runMem0 reasoning prompt, gpt-4.1-mini, Mem0’s lenient grader94.692.679.291.992.7
Mem0 Platform v3May 2026, own 2026 grader94.695.482.392.592.5
Supermemorygrader not listed, no per-category breakdown published––––77.1

Meko rows are our own runs; the others are each provider’s published numbers. Redis has published no LoCoMo result.

Learn more about the benchmarks
Up to 90%input token savings
97.9%LoCoMo session recall: the session holding the answer is in the top 50 results
91.8%LoCoMo single-hop judge accuracy with the standard paper grader (run 2)
Context Engine

The self-learning context engine for agentic apps

The self-learning context engine for agentic apps
Why a context engine

More context should mean better answers. It doesn’t.

01

The retrieval quality paradox

Too little context and the model guesses. Too much and the right facts are buried while cost climbs.

02

Context silos

Every agent and harness keeps its own partial, stale context and relearns what another agent already knows.

03

A stitched-together stack

Database, vector store, RAG, memory, conversation and artifact stores, each tuned apart, with no shared reranking or provenance.

How it works

One datapack. Every agent, tool and teammate.

A datapack replaces the database, vector store, RAG pipeline, memory layer and conversation store you would otherwise stitch together.

Claude
Cursor
Codex
Custom agents
Any harness
Datapack APIs: MCP · REST API · SQL
Conversationsverbatim, auto-tiered
Memoryprivate shared
Knowledgehybrid search
Decision tracesreasoning + tokens
Artifactsversioned outputs
PostgresStructured & Unstructured Data
YugabyteDB: Postgres · Vector · Graph · Automated RAG Pipeline · Document
Self-learning

Context quality, measured on every search.

When recall is weak, Meko says why and what is missing, then closes the gap. Every cycle starts smarter.

1

Search

Agents and people query the datapack each turn.

2

Score

Every result gets a 0 to 1 relevance score.

3

Explain

Weak scores are analyzed for what fell short.

4

Ask

Gaps become a concrete list of missing information.

5

Ingest

Sources fill the gap. The next search starts smarter.

context_search(“deploy checklist”)relevance 0.31Top hit is a log of an old request, not a fact. Missing: the current release runbook.
Use cases

What teams build on Meko

Featured

ADLC and software factories

Shared context for software projects in flight, with development standards layered at the organization, business-unit and project level. Every coding agent works from the same datapack.

Meko for ADLC
org › line-of-business › project
org/        coding standards, security rules
  lob/      service conventions, review checklists
    project/  decisions, open issues, agent learnings

Agentic support

Every answer adds to a shared context, so support agents stop relearning solved problems.

Agentic marketing

Campaigns built with full customer context, in hours instead of months.

Solutions blueprints

Field teams carry firm-wide best practices into each client project.

Fraud detection

Memory and context that help stop account takeover before it happens.

Pricing

Start free. Scale when you need to.

Compare plans

Free

$0

10 datapacks, 100 MB knowledge, community support. No credit card.

Standard

$25/mo

50 datapacks, 5 GB knowledge, 50 MB files, email support.

Pro

Custom

Unlimited datapacks, team sharing with roles, SSO, priority support.

FAQ

Questions teams ask first

What is Meko?
Meko is the universal, self-learning context engine for agentic applications. It gives agents and people one governed place for memory, conversations, knowledge and decision traces, delivered over a remote MCP endpoint. Learn more about Meko.
How is Meko different from the memory built into Claude, ChatGPT or Gemini?
Built-in memory lives inside one vendor’s product and one agent’s context. Meko sits outside any single model or tool, so the same context is shared across Claude, Cursor, custom agents and your teammates, and you decide what gets shared and with whom.
How do I connect to Meko?
Meko is a remote MCP server: point your MCP client at the Meko endpoint and sign in. Connection steps for Claude, Cursor, Codex and other clients are in the Meko docs.
Is Meko free?
Yes. The Free plan needs no credit card. Standard is $25/month for more datapacks, storage and email support, and Pro adds custom limits, SSO and priority support. See pricing or contact our team.