Enterprise · governed private memory
Keep what your agents learn.Govern who reuses it.
Enterprises are deploying agents across engineering, support, finance, data and operations. memco captures what those agents learn, curates what matters, and gives future agents the shortcut your organisation already paid to discover — inside boundaries you govern.
- deployment
- saas · vpc · on-prem · regional residency · audit trail
01The missing layer
You have the data. Not the learning.
Wikis, tickets, dashboards and data lakes record what happened. None of them capture what an agent programme learns while working: permissions quirks, workflow exceptions, approval paths, the fix that finally worked. That layer either compounds — or every team, vendor and agent rediscovers it alone.
- records of workwhat you already run
- Confluence
- SharePoint
- Notion
- Google Drive
- Box
- Dropbox
- Jira
- ServiceNow
- Zendesk
- Asana
- Trello
- Linear
- Slack
- Microsoft Teams
- Zoom
- Salesforce
- HubSpot
- Intercom
- Snowflake
- Databricks
- Google BigQuery
- Looker
- Power BI
- Tableau
- Datadog
- Splunk
- SAP
- Workday
- Oracle
- Okta
- GitHub
- GitLab
- Miro
- Figma
- Knowledge bases and wikis — written for humans, stale by design.
- Tickets, threads, retros — where insights go to be buried.
- Data platforms — facts about the business, not judgement about the work.
- learning from workwhat memco adds
- Captured insights from real agent runs, with provenance.
- Governed by default through trust and validation rules, with human review where you require it.
- Governed reuse: the next relevant agent starts ahead, inside your boundary.
02One governed workflow
From agent run to reusable memory.
memco sits beneath your agent stack. Not another chatbot, IDE or platform — the compounding layer underneath them.
- 01 · eng · support · finance · opsAgents work
the tools teams already use
- 02 · fixes · corrections · exceptionsmemco captures
with full provenance
- 03 · dedupe · score · decay · scopeMemory is curated
no taxonomies — signal from use
- 04 · permissioned · auditedGoverned memory
03The hard boundary
Company memory stays company memory.
Networks are logically isolated in storage and ranked separately at retrieval time. Nothing your agents learn contributes to any pool outside your organisation — there is no blended commons, and memco does not read your tenant.
- 01Private memory pools
Per-team, per-repo, per-org or per-deployment. Sharing across boundaries is opt-in and explicit.
- 02RBAC & permissioned sharing
Roles and permissions down to an insight. Promote, scope or revoke knowledge as a control-plane action.
- 03Provenance
Every insight traces back to the run, agent, repo and human correction that produced it.
- 04Decay
Stale conventions, outdated fixes and obsolete decisions lose weight before they mislead future agents.
- 05Audit trail
Every read, write, promotion and revocation is logged and exportable for compliance.
- 06Retirement
Memory can be edited, deduplicated and retired — deletion is a control, not a support ticket.
- 01 · sso · rbac · auditYour organisation
- 02 · validation rulesTeam & programme scopes
- 03Private memory
tenant-isolated · never trains models
04Provenance & audit
Trust is auditable, not asserted.
Every entry knows where it came from. Every action on it is logged.
Card data never transits the analytics VPC. Route exports through the tokenised gateway — see incident 2209.
audit log · excerpt
12:04:11 read constraint/pci-boundary agent · finance-07 12:04:12 serve rank 1 · trust .91 14:22:40 promote team → org user · j.okafor 14:22:40 log provenance chain attached
- exportable
- siem
05Deployment & residency
Runs where your boundary is.
Confirmed options only — and the same MCP and SDK interface in every one of them.
- option aManaged tenant
memco’s managed cloud, logically isolated: every read and write is scoped to your networks, with role-based access. Hours to deploy.
- option bYour VPC
The memco instance runs inside your cloud boundary. Regional residency options.
- option cOn-prem
Entirely behind your firewall. Org memory never leaves customer-controlled infrastructure.
- SOC 2 Type II
- GDPR-ready · DPA available
- SSO / SAML
- SCIM · roadmap
- UK / EU residency
06Pathfinder
A pilot with a success criterion, not a platform programme.
Start with one team, one workflow and one metric. The activation event we design for: an insight contributed on one task and reused on a second — by a different person, in a different tool.
- Weeks 1–2
One team connected · baseline measured: repeated discovery, token use, review loops
- Weeks 3–6
Memory accumulating · validation rules and scopes configured with your platform team
- Decision
Reuse rate, cost per task and audit posture reviewed against the criterion you set
Scope a pilot
Tell us the team, the workflow and the metric.
For programmes of a hundred engineers or more, regulated or procurement-heavy, with VPC or on-prem requirements, residency constraints or audit obligations. A founder reads every request and replies within two working days with a pilot shape. Under a hundred engineers and self-serve? Start a team on the free plan instead.
07Security & data flow
Asked by every security review.
Does our memory ever reach a public pool?
No. There is no public pool. Organisation memory is isolated per tenant, never aggregated, and nothing your agents learn is shared outside your organisation.
boundary · org ≠ public
What is your SOC 2 status, exactly?
SOC 2 Type II, independently audited. The report is available to customers and prospects under NDA.
compliance · type II
Where can it run?
Managed cloud, your VPC, or on-prem. Regional residency options. The interface is identical in every deployment.
deployment · saas · vpc · on-prem
Do you train on our data?
No. memco does not train on your code, prompts or completions. Memory belongs to the tenant.
data · tenant-owned
How does deletion work?
Entries can be retired or deleted at the control plane, with the action logged. Export is supported — offboarding does not strand your memory.
lifecycle · delete · export