Technical overview · Memco v1 · MCP & CLI

Shared memory,
made operational.

Memco Shared Memory is a memory layer for AI agents. It sits above any one agent — Claude Code, Cursor, ChatGPT — and gives every session the accumulated lessons of every session that came before it. Private by default; nothing you write leaves your workspace; curation runs without a human in the loop.

Start with

Your own private space. Bring any agent — what you solve once, your next session already knows, whichever tool you open it in.

01System overview

One memory. Every agent.

Multiple agents — different tools, different users — connect to a single Memco instance. They read via search, contribute via create and enrich, signal via feedback. Memco handles retrieval and curation; agents never coordinate with each other directly.

behind your firewall

Claude Code

user · alice

Cursor

user · ben

Windsurf

user · carla

+ n more

agents · ci · scripts

mcp · cli

Memco instance

continuous · autonomous

Hybrid retrievalvector + bm25 + trust
Trust modellingbayesian · decays
Memory opssynthesise · dedupe · prune
read · write

Org memory

private · default

your team’s knowledge · never leaves the tenant

tools · search · create_memory · enrich_memory · share_feedbackauth · sso · tenant isolation

fig. 01 — agents on the left, one memco instance, two memory scopes

Four agent nodes inside your firewall — Claude Code, Cursor, Windsurf and more agents, CI and scripts — connect over MCP and CLI to a single Memco instance running hybrid retrieval, trust modelling and memory ops. Memco reads and writes one scope: your organisation’s private memory.

02Agent interface

Core calls. Stateless on the agent side.

Agents talk to Memco over MCP — or, for non-interactive use, a CLI. The full surface is eight tools; the core read–contribute–feedback cycle is four.

01 · readat task start, on errors

search(query, context)

Retrieve relevant memories

Searches your organisation’s memory. Ranked by a hybrid of vector similarity, full-text match and trust-weighted evidence — not similarity alone.

hybrid retrieval · vector + bm25 + trust

02 · writeon discovery

create_memory(content)

Store a new insight

Triggered organically when an agent discovers an undocumented behaviour, a workaround, a convention. No prompt required. Enters the store with an initial trust score.

organic · a side-effect of work

03 · writeon outcome

enrich_memory(id, content)

Refine an existing memory

Adds context to a memory the agent has already read — for example, attaching the outcome of a task to the rule that informed it. The primary path for in-session refinement.

refinement · keeps memories current in place

04 · signalafter retrieval

share_feedback(id, score)

Was the memory useful?

Positive and negative scores both feed the Bayesian trust model. The closing edge of the loop — feedback from one session shapes ranking for every session after it.

closes the loop · evidence updates trust

03Trust modelling

Ranked by evidence, not similarity alone.

Every memory carries a trust score backed by a Bayesian distribution — not a scalar. Each retrieval, each relevance signal, each enrichment updates that distribution.

01

Bayesian, not scalar

Trust is a probability distribution. Each piece of evidence updates the posterior; confidence reflects both the level and the weight of accumulated signal.

02

Decays without signal

Memories that stop being retrieved or stop earning positive feedback gradually lose confidence. Nothing sits on the shelf forever because it once worked.

03

Exploration, not just exploitation

Retrieval samples from the distribution. High-confidence memories surface reliably; lower-evidence memories are selectively included so they can earn signal.

04

Server-side only

Trust is operational. The agent never sees scores and the user never sees them in chat. They live in the Memco dashboard, where they belong.

trust distribution · per memory

 trust →      0        .25       .50       .75        1
 ─────────────┼─────────┼─────────┼─────────┼─────────┼

 new          ▁▂▃▃▃▃▂▁
              wide · low weight

 emerging               ▂▄▆▆▄▂
                        narrowing

 validated                            ▂▆█▆▂
                                      tall · heavy
A memory's confidence is a distribution that narrows with evidence.Diagram of three trust distributions on a 0 to 1 scale: a new memory is wide, an emerging memory narrower, a validated memory a tall narrow peak near 0.9.

04Knowledge lifecycle

Memory ops — autonomous end to end.

From the moment an agent contributes, the curation loop takes over. No reviewer, no approval queue. Operators run continuously and write back into the same store the next search reads from.

queue · prior · index

Ingest

create or enrich lands, scored and indexed

vector · bm25 · trust · sampling

Retrieve

low-evidence memories get their chance

synthesise · dedupe · prune · safety

Curate

operators run continuously

posterior · re-rank

Update

every share_feedback re-scores

fig. 02 — the loop never breaks

05Memory scopes

Private by default. Private at every size.

Scopes are isolated at the storage layer and ranked separately at retrieval time. Nothing written in one reaches another.

● default · always onisolation · per-tenant

Organisation memory

Everything your agents contribute lands here, and it never leaves your tenant. Other organisations cannot read it, and Memco does not read it. Not shared with other customers, not aggregated, not used for model training.

all four calls operate against it by default

● always on · per organisationboundary · a firewall you control

Nothing leaves

There is no pool your memory drains into. What your agents learn stays inside the scope that wrote it — a team, or an organisation — and is never aggregated with anyone else’s.

no setting to turn on

06Portability

Memory should outlive every model and tool.

Memco is MCP-native and model-agnostic. The layer underneath can change; the memory remains yours.

Integration evidence — the stack Memco runs against today

Models
ClaudeGPTGeminiLlamaDeepSeekQwen
Agents
Claude CodeCodexCopilotClaude DesktopChatGPTGrok
IDEs
CursorWindsurfZedJetBrainsVS Code

07Deployment

Two ways to run it. Same interface.

Both models maintain full isolation of organisation memory and expose the same MCP and CLI surface. Migration is a config flip, not a rewrite.

option a · fully managed

Private tenant on Memco.

A dedicated Memco instance in Memco’s cloud, isolated at the data and compute level from any other tenant. Agents connect via secure HTTPS from behind your firewall.

hostingMemco cloud
org memoryper-tenant isolated
time to deployhours
option b · self-hosted

On-premises, in your infra.

A private Memco instance runs entirely behind your firewall. All organisation memory stays on customer-controlled infrastructure. Nothing crosses the firewall.

hostingcustomer-controlled
org memorystays on-prem
interfaceidentical · mcp & cli

08Empirical results

Compounding, not linear.

On controlled SWE-bench runs the first task is a cold start; every task after it benefits from accumulated knowledge. The variance reduction is, in our view, the most important number.

Memco controlled evaluation· claim ledger · pending reconciliation

−40%

LLM costs at steady state

−34%

Wall-clock time per task

−31%

Agent steps to completion

½×

Outcome variance — more predictable

swe-bench · steady state · memco evaluation · mar 2026

09Setup path

Running in one session.

Install the CLI, authenticate, initialise a scope. The two-agent tutorial on Getting started shows the first contributed and reused lesson.

spark — installnpm · cli · v1.x
$ npm install -g @memco/spark
spark login
spark init

10Personal memory

Your preferences. Every agent.

Memco Shared Memory is for your team. Personal Memory is for you — your preferences, working style, and context, stored once and available to every agent you use.

Switch between Claude, ChatGPT, Cursor, or any other MCP-compatible client and your agents already know how you work. No re-explaining, no per-tool configuration files, no context lost when you change tools.

Personal Memory is private, portable, and fully under your control. View, edit, and delete any memory through the Memco dashboard. It runs alongside Shared Memory on the same platform — personal context for you, institutional knowledge for your team.

Get started with Personal Memory

11Product FAQ

The technical questions, answered.

See it on your own repos.

Start a team trialTry Memco free

the loop

benchmarks · product · research

A short dispatch on shared memory for AI agents — the numbers behind the product, what we're shipping, and the research we're reading. No filler.

Unsubscribe anytime · no spam