Rules

Rules are persistent instructions that apply to your Qualia sessions. Unlike one-off prompts, rules are always active (or invoked on demand), ensuring consistent AI behavior across all your work.

​

What are rules?

A rule is a set of instructions stored as a file:

• Name: Identifier used to reference the rule
• Description: What the rule does (optional)
• Body: The actual instructions (up to 20,000 characters)
• Type: When the rule applies

On local workspaces, rules live as files in Qualia’s configuration directory. On Qualia Cloud, rules are stored with your account instead. Either way, they persist across sessions.

​

Rule types

​

Always rules

Applied automatically to every conversation:

• Loaded at the start of each generation
• Combined with other always rules
• No action needed to activate

Use always rules for:

• Coding style preferences (“Always use type hints”)
• Domain conventions (“Use SI units for all measurements”)
• Safety guidelines (“Never delete files without confirmation”)

​

Manual rules

Invoked explicitly when needed:

• Not automatically loaded
• Activated with @rule:rule-name in chat
• Useful for context-specific guidance

Use manual rules for:

• Workspace-specific conventions
• Specialized workflows
• Instructions that only apply sometimes

​

Pipelines

Pipelines use saved instructions to produce an artifact in a workspace. In Customize agent → Pipelines, select a pipeline and click Run now to start a new run. This works in local workspaces and Qualia Cloud. Each run opens its own chat and stores its artifact separately. In a shared workspace, pipeline definitions and schedules belong to the person who creates them. Other collaborators can see the resulting workspace files and run chats, but manage their own pipeline definitions and schedules. Pipelines with the same name created by different people have separate run histories. You can also ask in chat: “Make a pipeline that downloads Hacker News data and visualizes current trends.” The agent investigates the sources and establishes the requirements first, asking you about missing decisions when needed. It then saves reusable instructions and continues in the same chat as run 1, keeping the context from your conversation. The first run builds its report, dashboard, or presentation in its own output directory. Existing exploratory files stay where they are. Later runs follow the saved instructions, write separate outputs, and can read previous results to explain what changed.

​

Scheduled runs

On Qualia Cloud workspaces, pipelines can also run on a recurring schedule. Open the pipeline in Customize agent → Pipelines, click the pencil (Edit pipeline), and use the Schedule section:

• Frequency — every N minutes, hourly, daily, weekly (choose the weekdays), or monthly (choose the day of month), at a time of day.
• Timezone — schedules are evaluated on their own wall clock, so “9 AM in New York” stays 9 AM across daylight-saving changes.
• Ends — never, on a date, or after a set number of runs. A schedule that reaches its end turns itself off; skipped occurrences don’t count against a run limit.
• A plain-English summary and the next few run times appear as you edit, so you can confirm the schedule before saving. Picking the Custom cron frequency lets you write the cron expression directly.

The pipeline list shows each schedule’s next run time, and the schedule can be paused or removed at any time. A scheduled run starts even while you are working in the workspace — it opens as a new chat session like any other run. Scheduled runs are fully autonomous: the agent never stops to ask for confirmation and proceeds on its own best judgment, since nobody may be around to answer. If the previous run is still active when the schedule fires, that occurrence is skipped rather than queued (the pipeline shows “waiting for the current run to finish”), and downtime never replays missed runs: at most one run starts when the schedule comes back due. Scheduled runs are billed to the workspace owner and pause automatically when the account’s usage cap is reached. In local workspaces, runs are started manually. Each scheduled run checks that its creator still has Owner or Editor access. Losing Editor access prevents future scheduled runs and stops that person’s running agents and jobs.

​

Rule scope

Every rule is either global or scoped to the current workspace:

• Global rules apply in every workspace.
• Workspace rules apply only in the current workspace.

Pick the scope when creating a rule, and change it later by editing the rule.

​

File path rules

Besides free-form rules, you can add a file path rule (click New file path in the Rules tab). It points the agent at a folder of code on disk: when the rule applies, the agent assumes imports in your notebooks resolve from that path. File path rules are always active, and like other rules they can be global or scoped to a workspace. A global file path applies in every workspace, so use one for shared code — and make sure the path exists on the machine running the backend.

​

Creating rules

1. Open Customize agent and select Rules
2. Click Create Rule
3. Enter:
   • Name: Lowercase with hyphens (e.g., python-style)
   • Description: Brief explanation
   • Body: The instructions
   • Type: Always or Manual

​

Using manual rules

Invoke a manual rule by typing @rule: followed by the rule name in your chat message:

“@rule:data-analysis Analyze this dataset and find anomalies”

The rule’s instructions are included in that message’s context.

​

Workspace rules

Qualia also recognizes rules files in your workspace root:

• CLAUDE.md: Instructions in Claude’s standard format
• AGENTS.md: Agent-specific guidance

If present, these are included automatically like always rules. This lets you version-control workspace-specific instructions alongside your code.

​

Importing rules

Import rules from other tools. These options read directories on the machine your workspace runs on, so they appear for local workspaces only:

​

From Cursor

1. Open Customize agent and select Rules
2. Click Import from Cursor
3. Rules from ~/.cursor/rules/ are imported

Cursor rules (.mdc files) include frontmatter with globs for file-specific matching. Imported rules preserve this metadata.

​

From Claude Code

1. Click Import from Claude Code
2. Rules from ~/.claude/rules/ are imported

​

Re-importing

Importing is a copy, so a rule you already imported is left alone the next time you import from the same source. If the source has changed since, Qualia lists the rules it would replace and asks before touching them: Keep mine leaves your copies as they are, and Update from Cursor / Update from Claude Code replaces each listed copy with the current source version, discarding any edits you made to it in Qualia. Rules that are new to Qualia are imported either way. Workspace-scoped rules and rules named after a built-in are skipped during re-import.

​

From a file

1. Click Import from file…
2. Enter the full path to a file (such as a CLAUDE.md) on the machine running the backend
3. Its contents become a new always rule

Use this to bring in an existing system prompt or CLAUDE.md that lives outside the standard ~/.cursor/rules or ~/.claude/rules locations. Importing from a file always creates a new rule: if the name is already taken it is skipped rather than replaced, so rename the existing rule first if you want the file’s contents to win.

​

Rule organization

For complex workspaces, consider:

• One always rule for global preferences
• Manual rules for different contexts (data analysis, model training, visualization)
• Workspace rules (CLAUDE.md) for workspace-specific conventions

​

Viewing rule files

On local workspaces, click Open Rules folder to open the rules directory in your file manager. Each rule is stored in its own folder with a RULE.md file. You can edit rules directly on disk if you prefer — changes are picked up automatically. (Cloud workspaces store rules with your account, so there is no folder to reveal; edit rules from the Rules tab instead.)

​

Best practices

Be specific: Vague rules like “write good code” don’t help. Specific rules like “always use TQDM in for loops” and “add docstrings to all functions” give clear guidance. Keep rules focused: One rule per concern. A rule about Python style shouldn’t also cover data visualization preferences. Test rules: After creating a rule, try a few prompts to verify it’s being followed. Review periodically: As your workflow evolves, update rules to match. Outdated rules can cause friction.