Spools

Spools: create your first team and agent

Create a Spools team and agent, customise role instructions and variables, choose a model, and enable only the tools you need.

Before you start

Spools is Neurothread’s agent area. Its actual account-menu entry and page heading are Agents, with a Beta badge. Sign in and open the avatar menu to check whether the entry is available to you.

Spools is in early access: the Agents entry only appears for accounts that have been given access. Owning or administering your account does not enable it, and neither does upgrading your plan. Once you have access, the agent computer is a separate capability of eligible paid plans.

Prepare a small task and non-sensitive company context. Teams group your agents; they are not a way to invite or collaborate with other people. Review tool settings before sending the first message: role templates can enable more tools than an agent created from scratch.

When to use this

Use an agent when you want reusable instructions, a continuing agent conversation, company and team memory, or scheduled messages. For a one-off question with visual branching or parallel model comparison, the ordinary Canvas or Chat may be simpler.

Start with one team and one clearly defined role. A role template provides editable instructions and named variables; it does not guarantee specialist accuracy or automatically coordinate a workforce of agents.

Steps

Open the account menu → Agents. On an empty account, select Create your first team; otherwise use New team beside Teams.
In New team, enter a Name, select a Type, and click Create team. Types include Marketing, Engineering, Operations, Finance, Sales, Support, Leadership, and Other.
Optionally fill in Company memory and the team’s menu → Team memory before creating the agent. Matching facts can fill a role template’s variables at creation. Use only information appropriate for the agent to receive.
Select New agent, using the person-add control beside the intended team. Under Start from a role, choose a listed template, such as Marketing lead or Software engineer, or choose From scratch. Enter a name, verify Team, and select Create agent. When no team was preselected, a role may suggest a team with a matching type; check the final selection.
Open the agent’s conversation and the settings control → Agent settings. Review Instructions and the existing Variables values. Words such as {{company}} in instructions are replaced by that agent’s saved variable values. The dialog edits existing variable values; it does not offer an add-variable control.
Under Model, choose Automatic or Choose a model. Automatic currently resolves to the platform’s configured default agent model; it is not a promise that Jev dynamically chooses the best model for each task. To control the choice yourself, select an available model in the picker.
Under Tools, review Web search, Connected apps (MCP), Neurothread canvas (reads your workspaces and saves documents there), and Own computer (browser, terminal, files). When Skills is available, Use skills (your automatic skills, loaded when a task matches one) lets agents load eligible automatic skills. Turn off unnecessary capabilities, then Save. Computer is locked without an eligible paid plan; model tool support and service availability also matter.
Send a bounded first task in the agent chat. Review the output and tool activity before expanding permissions or creating routines. Use the team’s menu to rename it; use Team in agent settings to move an agent, understanding that this changes its team-memory scope.

Worked example

Illustrative exercise, not an executed run: create a Marketing team and a Marketing lead named “Draft reviewer”. Set safe company facts first, then review the copied company, product, audience, and tone variables. Disable MCP and Computer if you only need writing help.

Ask: “Using the product and audience in your instructions, draft two short launch-post variants. Do not publish or send anything. List assumptions for me to check.” Read both variants and correct any invented claims before using them.

What happens next

The agent keeps its name, team, original editable instructions, variable values, model mode, and tool settings. Creation saves configuration version 1; changes save further configuration versions. The current settings dialog does not provide a version-history browser or a Restore button. Keep your own copy before major edits instead of assuming an in-app rollback exists.

Template variables are copied from memory at creation, not continuously synchronised. Later memory changes do not automatically rewrite those saved values. Each subsequent task can also receive the current company and own-team memory. Moving an agent changes which team memory it receives.

Delete agent removes its conversation and routines along with its saved configurations. Delete team also removes its agents, routines, and team memory. These are not ways to undo effects in external services; review the confirmation carefully and avoid deleting work you need to retain.

Cost and limitations

Spools uses the account’s existing plan and credit allowance. Model turns, additional tool rounds, searches, and computer use can contribute to costs; enabling a tool is not an unlimited-use allowance. Plans impose team, agent, and routine limits. Treat the app’s limit message as authoritative rather than assuming a fixed quota will remain unchanged.

For BYOK model billing, save a key for the selected model’s provider in Settings → API keys, have a plan that permits BYOK, and use Choose a model to select a compatible provider model. The worker uses the same own-key lookup as ordinary answers; there is no separate key field in Agent settings. Neurothread house models always use platform billing. Automatic defaults to Neurothread Default, so adding a provider key does not by itself make automatic agent turns BYOK. Computer and applicable search charges remain credit-based; search is disabled when BYOK is active but the account has no credits left.

Instructions are limited to 20,000 characters; team and agent names to 80. A chosen model must be available, and tools require a model with tool-call support. The default settings for an agent from scratch enable web search, canvas, and skills, while role templates can also enable MCP and Computer where eligible. Integrated memory, canvas, and computer tools can run automatically; MCP approval rules are different. Do not assume every action that changes data will ask first.

Troubleshooting

No Agents entry: early access may not be enabled for your account. Administering your account or upgrading your subscription does not grant early access.
Cannot create an agent: create a team first, enter a non-empty name, and check the displayed plan limit.
Company details are missing or outdated: review the template’s expected memory facts and the agent’s saved Variables. A later memory edit does not refill existing variables.
Automatic seems to use the same model: it currently uses the configured agent default. Use Choose a model for an explicit choice.
The tool does not appear or run: check the agent switch, model tool-call support, and the relevant connection, Skills access, or computer eligibility.
Deletion says to wait: a task may still be queued or running. Inspect the agent conversation and use Stop if appropriate; stopping is not an external rollback.