Spools
Spools memory: company facts and team notes
Manage your company and team memory, fill role-template variables, review agent notes, and avoid overwriting concurrent changes.
Before you start
Section titled “Before you start”Open the account menu → Agents, the actual entry for Spools. Spools is in early access and only appears for accounts that have been given access; a paid subscription does not enable it.
Spools has two memory scopes, both owned by your user account:
- Company memory is shared by all of your agent teams. “Company” does not mean other people or accounts can read or edit it.
- Team memory is supplied only to your agents in that team, not to your other teams.
These stores are separate from personal Neurothread memory in Settings → Memory. The personal-memory switch, sensitive-section rules, and earlier-summary restoration controls are not controls for Spools memory. Do not assume that turning personal memory off clears or disables company and team memory.
When to use this
Section titled “When to use this”Put stable, non-sensitive company context in Company memory: a product description, brand tone, or company name. Put team-specific goals, audience, or technical stack in Team memory. Keep task-specific requests in the agent’s chat and role-specific behaviour in Agent settings → Instructions.
Use Facts for short named values that templates can read. Use Notes for decisions and context that do not fit a single value. This is working context for models, not a password manager, secret vault, authoritative database, or versioned document archive. Avoid credentials, confidential customer records, and unnecessary personal information.
- In Agents, open Company memory above the team list for information that every one of your teams should receive.
- Review the existing Facts and Notes before editing. Keep a safe copy if you need to preserve the exact current text. Do not delete notes just because you did not type them: an agent may have added them.
- Select Add fact. Enter a short Name and Value. Names use lowercase letters, numbers, and underscores and must begin with a letter, such as
companyortarget_audience. The name field normalises accents and spaces; inspect the resulting key rather than assuming it retained your label. - If offered, use a suggested fact under The role templates use:. These suggestions come from the templates’ expected source keys. Filling an unrelated key with a similar label will not necessarily fill the intended variable.
- Edit Notes to retain useful context and remove obsolete or inaccurate material. To remove a fact, use its delete control; rows with no non-empty value are omitted on save. Click Save only after reviewing the complete contents of the dialog.
- Open a team’s menu → Team memory to edit its separate facts and notes. Check the title Memory of [team name] so you do not put team-only context in Company memory by mistake.
- Before creating a role-based agent, fill the facts that its template uses. For Marketing lead, company, product, and tone come from Company memory; audience comes from its team’s facts. After creation, open Agent settings → Variables and check the copied values.
- In the agent chat, explicitly ask it to remember a short, appropriate fact and specify “for this team” or “for all my teams”. If it uses its
remembertool, inspect the saved note in the relevant memory editor. A conversational claim that it remembers something is not a substitute for checking the stored note.
Worked example
Section titled “Worked example”Illustrative exercise, not an executed run: save Company facts company = “Acme Demo”, product = “A fictional planning app”, and tone = “Clear and practical”. In the Marketing team’s memory, save audience = “Small design studios”. Create a Marketing lead and check its corresponding variable values in settings.
Later ask: “Remember for this team that launch drafts must include an assumptions list.” Review the Marketing Notes for that entry. If the audience changes, update both the appropriate memory fact and any existing agent variable that still contains the old audience. Do not expect changing the fact alone to rewrite the saved variable.
What happens next
Section titled “What happens next”The agent’s instructions are rendered with its saved variable values, then its company and own-team memory are added as context. Company and team memory are not filtered by the personal-memory sensitive-topic switch. Treat everything in these stores as potentially supplied to the agent’s model when it works.
Template autofill happens at creation. Editing memory later changes the memory context, but does not continuously refill variables in existing agents. Moving an agent to another team changes the team memory it receives without automatically replacing its saved variable values.
The integrated remember tool automatically appends a short note to Company or Team Notes; it does not update the named Facts table. It is available to tool-capable agent runs independently of the MCP, canvas, computer and skills switches. An agent can choose to call it without an explicit “remember” request, and there is no separate Spools-memory off switch in Agent settings. Review notes for accuracy and appropriate scope. Neurothread does not keep earlier versions of the memory that you can restore from the app; a version counter is not a recoverable archive.
Cost and limitations
Section titled “Cost and limitations”Editing and saving this memory does not itself generate a model answer. Using it in tasks adds context that can affect model token costs. Asking an agent to remember something is a model task and can incur normal run costs.
Each memory scope supports up to 50 facts, values up to 500 characters, and Notes up to 20,000 characters. Keys are at most 40 characters. An automatic remembered note is limited to 500 characters and is refused when Notes is full. Keep context concise; filling the maximum does not make every model able to use it effectively.
Saving replaces the complete Facts and Notes payload loaded in the dialog; it is not a merge. There is no conflict-checking control in this editor. An agent can append a note while you have the editor open, or another tab can save newer values. Saving your older copy can overwrite those changes. Avoid editing while agents are adding notes; close and reopen to load the latest content, reconcile it with your intended edits, and then save. Keep a copy of your edits before closing if necessary. Never assume concurrent updates are automatically protected.
Concurrent agent note appends also read and replace the Notes text without a conflict check. Review the stored notes after simultaneous tasks rather than assuming every append survived.
Troubleshooting
Section titled “Troubleshooting”- A fact did not fill the template: check its exact source key and scope. Company
audienceis not the Marketing template’s teamaudience. Autofill affects new agents only. - An existing agent still names the old company: check Variables and Instructions, not just memory. Copied variable values can remain stale.
- Save is disabled: wait for memory to load and correct invalid names. Use an initial lowercase letter and only letters, numbers, or underscores.
- A fact disappeared after saving: a blank value is omitted; duplicate names become one map entry. Use a distinct key for each fact.
- A remembered note is missing: confirm that the tool actually ran, check the selected scope, and tidy Notes if it is full. The note is not stored as a new fact row.
- Recent notes disappeared after a manual save: an older dialog may have replaced newer contents. Do not assume a restore button exists; check a copy you retained and re-enter verified content.
- Personal memory is off but an agent knows company details: these are separate stores. Review Company and Team memory, saved variables, and instructions directly.
