memco for Knowledge Work
Every task should make
the next agent better.
One agent learns. The next starts ahead.
Knowledge work is where judgement is the asset.
- research
- client work
- operations
- recurring briefs
- decisions
- research
- “Vendor blog posts are leads, not evidence — two independent sources before a claim ships.”
- client work
- “State the recommendation first: this client reads nothing past the opening paragraph.”
- operations
- “Refunds over €500 need a second approver since the March policy change.”
- recurring briefs
- “The Friday brief fits one screen: decisions first, then asks, then calendar.”
- decisions
- “Record what was decided, why, and what it superseded — not just the outcome.”
- Today
- The task ships, the correction lands — and the judgement disappears into the chat.
- With memco
- It becomes a sourced, scoped insight the next agent reuses — across models and tools.
- Private by default · Scoped to the right work · Sourced to the original evidence · Owned by you
Live demo · the second task · Monday → Friday
See one correction on Monday improve a different agent’s work on Friday.
With memco
- Monday · agent 1 · client updateClaude Cowork
Drafts the weekly client update.
- Your correction
- “Lead with decisions and blockers. Separate confirmed facts from open assumptions.”
- memco · Knowledge Work · Private spaceWeekly client update structure
Lead with decisions and blockers; label assumptions explicitly.
- Source
- user correction · Claude Cowork
- Evidence
- accepted final draft
- Scope
- client reporting · private space
- Status
- trusted
- Reused
- Friday · ChatGPT · this demo
- Feedback
- positive · improves future retrieval
- Trusted
- user correction
- client reporting
- Friday · agent 2 · different accountChatGPT
Prepares the update for a different account.
- Applied before you had to repeat the correction · from “Weekly client update structure”
- 1. Decisions
- 2. Blockers
- 3. Open assumptions · labelled
- Later agent · Coding agents
- Claude Code · Codex CLI · Gemini CLI · GitHub Copilot CLI · Cline · Aider · goose · OpenHands · Amp · Devin · Jules · opencode
- Later agent · Editors & IDEs
- Cursor · Devin Desktop · Zed · VS Code · JetBrains Junie · Antigravity · Trae · Kiro · Continue
- Later agent · Assistants
- ChatGPT · Claude · Claude Cowork · Gemini · Microsoft 365 Copilot · Perplexity · Notion AI · OpenClaw · Hermes
- Connectors
- correction ▼ · relevant insight ▼
Without
- Monday · agent 1 · client updateClaude Cowork
Drafts the weekly client update.
- Your correction
- “Lead with decisions and blockers. Separate confirmed facts from open assumptions.”
- memco · not connected
The correction stays inside Monday's chat. Nothing carries it forward.
- Friday · agent 2 · different accountChatGPT
Prepares the update for a different account.
- Decisions buried mid-summary · your Monday correction never reached this agent
- 1. Progress summary
- 2. Activity log
- 3. Next steps
- Connectors
- correction ▼ · nothing retrieved ▼
Different task. Different agent. Same hard-won insight.
memco sits between your agents as the learning layer: one private, domain-aware memory, connected once over MCP.
Recognisable work
The work where judgement is the asset.
Four workflows, the same pattern: what happened, what memco learned, what the next agent does differently.
- knowledgeConsulting
“State the recommendation first — this client reads nothing past the opening paragraph.”
- before
- Every draft buried the recommendation under context
- correction
- Claude Cowork
- → next task
- Friday · Hermes drafts the steering-committee note that way
- knowledgeResearch
“Vendor blog posts are leads, not evidence. Two independent sources before a claim ships.”
- before
- A vendor claim slipped into a memo as evidence
- correction
- Claude
- → next task
- next scan · weak evidence flagged before the memo
- knowledgeOperations
“Refunds over €500 need a second approver since the March policy change.”
- before
- A refund went out without the second approver
- decision
- Hermes
- → next task
- new case · a different agent routes it right
- knowledgeAssistant work
“The Friday brief fits one screen: decisions first, then asks, then calendar.”
- before
- Every Friday started with the same formatting instructions
- preference
- Claude Cowork
- → next task
- every Friday since · no re-explaining
Evidence
Don’t take our word for it. Run the learning loop.
Memco controlled evaluation · not a customer deployment result
In a controlled banking evaluation, learning from corrections reached 2.6× the single-trial success of a static-RAG baseline. The memory transferred between Mistral Large and Claude Sonnet 5 in that setup.
Activation has one definition here: the first later task that visibly benefits from a prior insight. You measure it on your own work, not ours.
The example calls your own private memco space, so it needs a free signup first.
What memco keeps
Keep the judgment. Lose the chat sludge.
Chat history and retrieval preserve what was said and what is written down. Neither turns a correction into something a different agent can act on next week.
- keepsChat history
What was said
Held until the window closes, inside one tool. - keepsRetrieval and knowledge bases
What was written down
Documents and facts, reachable by many tools. - keepsmemco
What the work taught you
Corrections, accepted outcomes, failed paths, decisions and changed policies — sourced, scoped and reusable by a different agent later.
The expensive part is not storing context. It is paying to rediscover the same judgment.
How an insight becomes trusted
Intelligence compounds only when the insight is trusted.
A correction is not an insight yet. It becomes one when it has a source, a scope and enough evidence to be validated — and it stops being one when the world changes.
- 01 · correction or outcomeLearn
Something you fixed, decided or proved becomes a candidate insight.
- 02 · source · scope · evidenceGovern
memco attaches where it came from, where it applies and what makes it trusted.
- 03 · a different task, laterReuse
A future agent applies it; feedback updates or retires it.
feedback updates the insight — or retires it
Source, scope, status and retirement are visible objects on every insight — not fine print. Without them, memory does not compound. It becomes sludge.
- source
whose correction, which task
- scope
where the insight applies
- status
trusted · provisional
- retirement
old deploy advice · stale · retired
Ownership
Your agents can change. Your intelligence should not disappear with them.
Everything your agents learn lands in a space you own — private for you, shared when you invite your team.
- 01Provenance and evidence
Every insight carries its source: whose correction, which task, what evidence made it trusted.
- 02Scope and permissions
Insights apply where they belong — a client, a workflow, a team — not everywhere at once. Private by default; nothing is published anywhere outside your space.
- 03Contradiction and retirement
A conflicting insight surfaces for review instead of silently merging. Stale knowledge loses trust and is retired — struck through, not quietly deleted.
- 04Revocation and export
Disconnect a client, revoke access, take your insights with you. Change model, agent or interface without throwing away what use has taught your agents.
OAuth 2.0 · API keys for automation · encrypted in transit and at rest · regional and data-residency options subject to plan
One connection, separate domains
One MCP connection. Separate Coding and Knowledge Work stores. A repository fix never leaks into a client-reporting workflow.
Connect once
Bring the agents you already use.
memco is the learning layer between them, not another place to do the work. One secure MCP endpoint, no plugin per tool, nothing re-explained.
- Claude
- Cursor
- Codex
- GitHub Copilot
- Zed
- JetBrains
- Hermes
- OpenClaw
- Google Gemini
- Notion
- Windsurf
- Buzz
shared memory
model agnostic · IDE agnostic · tool agnostic
- custom connectorClaude / Cowork
- remote MCPHermes
- documentedClaude Code
- API keys availableAny MCP client
Connector availability and setup differ by client and plan
Plans
Start free. Let the insights accumulate.
Prove the learning loop on your own work first. Move to organisation-owned controls when it becomes production-critical.
| Row | FreeFree · up to four members | Team$50 per member per month | EnterpriseVPC and on-prem deployment options |
|---|---|---|---|
| Up to four members, you and your agents. | $50 per member per month. Unlimited networks, so a team can run a parent and children. | Unlimited networks, validation rules, policies, residency, and your own domains. | |
| Private space · both domains · MCP access | Org scopes · unlimited networks | VPC and on-prem deployment options | |
| Create your private space | Start a team | Scope a pathfinder |
- Full seat pricing on Pricing
Questions
Asked before connecting.
What is memco for Knowledge Work?
Knowledge work is the research, analysis, client delivery, decisions and recurring operations where judgement matters. memco for Knowledge Work is memco’s built-in domain for that work — it gives those insights the right semantics for processes, policies, stakeholders, terminology and procedures. It is not a separate product: memco is one product with two built-in domains, Coding and Knowledge Work.
Does memco read all my files and conversations?
No. memco operates through explicit MCP tools for search, contribution, enrichment and feedback. The connected agent decides what it asks memco to retrieve or save, under its own instructions and permissions. There is no background crawl of your files or chat history.
What if an insight is wrong or stale?
Agents give positive or negative feedback, and memco uses that evidence in retrieval and curation. An insight that stops earning signal loses trust and is retired — struck through, not quietly deleted. A contradicting insight surfaces for review rather than silently replacing the old one.
Can I use the same learning across agents and models?
Yes, wherever the client supports the remote MCP connection. They authenticate to the same private space and use the domains available to it. In the controlled evaluation below, memory learned on Mistral Large transferred to Claude Sonnet 5. Connector availability and setup can differ by client and plan.
What is free, and when do I need Team or Enterprise?
Free covers up to four members. Team adds organisation-owned learning, unlimited networks and production controls. Enterprise covers deployment, governance and regulated programmes. Current seat pricing is on the Pricing page.
Start with the next repeated insight
Make the next task the proof.
Connect one agent, correct one task and see whether the next related task starts ahead.
Free to start · No public sharing required · MCP setup in minutes