Turn what your people teach AI into shared company know-how.
Bring lessons from Claude Code.Claude Code.
OMS turns corrections and working practices into shared skills for your AI agents. Keep guidance current, control changes and make useful lessons available across teams.
Organisational memory
What is the OMS?
The Organisational Memory System manages the skills your AI agents use: reusable instructions for how your organisation works. It turns corrections into shared guidance, checks changes and keeps a history.
Turn one person’s useful lesson into guidance the whole organisation can build on.
Why it matters
Useful skills need more than a place to live.
A lesson in one conversation does not automatically become current guidance for the rest of your organisation.
Lessons stay local
One person finds a better way to do the work. Another team has to discover it again.
Instructions fall out of date
Your process changes, but an old skill keeps giving yesterday’s instructions.
Changes are hard to trace
Across copied files and different tools, it is difficult to see who changed the guidance and why.
Independent research
Keep useful guidance controlled, current, and shared.
Good skills can improve results.
+16.6
percentage points in average pass rate
SkillsBench found that curated skills raised average pass rates from 33.9% to 50.5% across 87 tasks and 18 model-harness configurations.
A useful method is worth making reusable across your organisation.
SkillsBench, version 4See it work
One correction. A lesson the next agent can use.
Illustrative example, not a live OMS session.
Follow one lesson to its next use
New session · same task
Agent: Supplier approval checked. Quote request prepared.
Rule followed
supplier_quote · skill file
Added ruleCheck the approved supplier register before requesting a quote.
This time, the agent checks supplier approval before preparing the quote request.
Governance
Control what changes. Know what changed.
A proposed change isn’t permission to distribute it.
Held for review · unpublished
Skip supplier approval when urgent.
Proposed change
Record the supplier approval reference before requesting a quote.
Procurement team
Skill file · user 1
Check supplier approval before requesting a quote.
Added ruleRecord the supplier approval reference before requesting a quote.
Skill file · user 2
Check supplier approval before requesting a quote.
Added ruleRecord the supplier approval reference before requesting a quote.
On configured refresh, Purchasing users receive the approved update.
Set who can approve changes
Use contributor trust and review policies to control how guidance changes.
Choose which teams receive guidance
Scope published skills to the teams that need them.
Trace each rule to its source
See the original correction, review decisions and published versions.
Why add OMS?
Make useful lessons travel further. Control what goes with them.
Sharing agent guidance can spread a better way of working, or spread a mistake. OMS helps you reuse useful lessons, check proposed changes before distribution and keep your organisation’s guidance independent of any one AI platform.
Stop paying for the same lesson.
Make useful corrections available to the next person’s agent, so your experts spend less time repeating guidance they have already established.
Reduce the risk of spreading harmful guidance.
Screen contributions for malicious instructions, check for conflicts and use review policies to control changes. Choose which teams receive the guidance, so an unchecked contribution does not automatically become everyone’s instruction.
Change platforms without starting over.
Keep your organisation’s guidance under your control and deliver it to compatible agents as your tools change.
Illustrative, not measured
One five-minute repeat correction avoided per person each week is 1,000 hours a year for a 250-person team.
Illustrative: 48 working weeks. At a £450 loaded day rate and a 7.5-hour day, that is £60,000 of staff time, before OMS and operating costs.
Estimate the value for your teamQuestions
Common questions
No. Conversation memory helps one agent carry on a conversation. OMS manages the rules your organisation works by, across people, teams, sessions and different agent tools.
Yes. Git gives you file sharing and version history. OMS adds a process for submitting lessons, checking duplicates and conflicts, replacing superseded guidance, tracing rules back to their source corrections and publishing to users automatically. For a small team with light usage, a hand-maintained skill library may be enough.
Your platform’s sharing and administration features remain useful. OMS adds a platform-agnostic workflow for capturing lessons, reconciling them with existing guidance and publishing changes across compatible tools. Whether you need it depends on the maintenance work your current setup leaves to your team.
No. Compatible agents see what skills are available and load the detail only when the task needs it. The long instructions sit in files the agent opens on demand.
After installation, OMS refreshes published skills for enrolled users and teams on a configured schedule. That refresh needs the right permissions and connectivity, and it applies to the next session rather than one already running.
You control which contributions can progress automatically and which require review. Agent learnings are suggestions, and locked constraints are protected by publication checks. These controls govern the guidance; they do not guarantee that a model will follow it.
Agents using already-published skill files can continue reading them without OMS. New updates and corrections wait until the connection returns. Agents requesting skills over a live connection need OMS to be available for those requests.
No. It does not assemble prompts. It provides skills through configured files or MCP, and the compatible agent decides when to load the relevant guidance.
No. Existing memory tools can continue to serve their purpose. OMS manages organisational guidance and can sit alongside them.
Contributions are screened for supported personal-data patterns and prompt injection before storage. Screening does not guarantee detection of every sensitive detail. Organisations’ rules are kept separate, and published skill files are generated from the managed guidance.
Get started
See OMS on the agents you already run
Thirty minutes. Bring a few corrections your people keep repeating, and we will show you what OMS does with them. We either walk you through the whole system, or help you install the free version and leave you to it.
Under 30 seconds. We will only email you about OMS.