QuickAgent - Guide

QuickAgent User Guide

QuickAgent is a local-first AI agent workspace. It turns a goal into a durable Run, limits that Run with a Workspace Profile, asks for review before sensitive actions, and keeps verified results in Artifacts.

Getting Started

Open Settings > AI, add a provider, and run its connection test.
Open Settings > Workspace Profiles. Choose which provider, tools, connections, file roots, Knowledge sources, and Memory scopes may be used.
Select New Chat, choose the profile, and confirm the provider destination shown above the conversation.
Describe an outcome. Add only the files, folders, media, Knowledge, or Memory needed for this Run.
Review the plan, approvals, Critical confirmation, and Data Send Preview when they appear.
Track work in the Timeline. Open the completed output from Artifacts and inspect the audit trail in History.

Tip: Start with a narrow Workspace Profile. Expand its tools or roots only when QuickAgent shows a specific unavailable reason.

Interface Overview

Section	What it is for
Chats	Conversations, provider destination, Context Tray, plans, approvals, and the active Run Timeline.
Runs	Every planned, running, paused, scheduled, failed, or completed task.
Approval Inbox	Pending action, Critical, and Data Send reviews that need your decision.
Artifacts	Verified files, reports, logs, media, tables, and operation manifests created by Runs.
Automations	Schedules, monitors, webhooks, app events, and Shortcut/App Intent triggers.
Skills	Reusable instructions with typed inputs, allowed tools, required connections, risk, and outputs.
Connections	Providers, browser profiles, file/VFS roots, SSH, MCP, HTTP/SaaS, home bridges, and notification channels.
Knowledge	Indexed folders, QuickNote content, web clips, PDFs, and Office sources.
History	Append-only safety decisions, tool actions, destinations, and timestamps.
Settings	Appearance, language, AI, tools, safety, Workspace Profiles, shortcuts, permissions, Help, and Pro.

Providers and Workspace Profiles

QuickAgent can use System AI when the operating system exposes a compatible capability, local providers such as Ollama, LM Studio, or DeskLLM, LAN or custom OpenAI-compatible endpoints, and configured cloud providers such as Anthropic or Gemini. Availability is determined by a live capability test; a provider name alone does not mean it is ready.

A Workspace Profile is the Run’s permission envelope. It snapshots the selected provider/model, allowed tools, connections, file roots, Knowledge sources, Memory scopes, safety overrides, and budget. A later configuration change makes the snapshot stale and requires review.

Destination rule: QuickAgent never silently switches from an on-device or local provider to cloud. The active destination remains visible in Chat and in every Data Send review.

Runs, Approvals, and Recovery

Run controls

Pause requests a cooperative stop at a durable checkpoint.
Resume continues the same Run and completed work.
Stop cancels provider and tool work. A stopped or uncertain external effect is not reported as success.
Retry is available only when QuickAgent can identify a safe retry boundary.

Safety levels

Level	Meaning
Safe	Read, inspect, or transform within the active profile.
Approve	Review a normal write or external action.
Confirm	Review a higher-impact or harder-to-reverse change.
Critical	Strong confirmation every time; rules never auto-approve it.

Approval shows the exact target, destination, proposed arguments, risk, and Undo status. If you edit an argument, QuickAgent validates and risk-checks the edited action again.

Uncertain outcome: after an interruption, check the target system first. Only choose “I checked — retry” when repeating the action is appropriate. “Resume” alone never repeats an uncertain external effect.

Multi-Agent Runs

Complex goals can use a planner, typed specialist workers, a reviewer, and a security reviewer. The Timeline shows each handoff, review, retry, fallback, and final-response boundary. Workers receive only the four-way intersection of Skill tools, Workspace Profile tools, role tools, and currently live tools.

Global and per-role budgets limit invocations, time, cost, network bytes, and tokens. When a budget is reached, QuickAgent pauses. It shows the current and proposed limits, and increases them only after explicit review. The approved budget and completed role history survive a restart.

Context, Knowledge, and AI Memory

Context Tray

Add files, folders, images, video frames, audio, voice notes, QuickNote content, Knowledge, or Workspace Memory. QuickAgent extracts or summarizes supported content and shows whether an attachment will be sent directly to a provider.

Knowledge

Knowledge sources are indexed separately from chat. A Run can search only sources allowed by its Workspace Profile, and citations retain their source identity.

Memory

Memory is deliberate, not automatic chat history. User, Workspace, Project, and Tool/Connection scopes are separate. Every create, edit, or delete request receives a fresh approval, and the approval preview shows the scope and exact proposed value.

Tools and Connections

The live catalog covers file/VFS operations; Office, PDF, and structured data; media and OCR; browser and computer use; development commands; SSH; MCP; HTTP and SaaS; smart-home bridges; notifications; schedules; monitors; and automation control. Connected MCP servers may add dynamic tools after schema validation and QuickAgent risk reassessment.

Settings > Tools shows why a tool is ready or disabled. Common reasons are a missing connection, operating-system permission, unsupported platform, Workspace Profile restriction, untested endpoint, or unavailable provider capability.

Computer use: screen capture and accessibility/input control are separate permissions. Keep the visible preview and emergency Stop available. Mobile platforms do not provide the same unrestricted desktop control.

Automations

Automations can start from time, file/folder, web, webhook, email, calendar, network, host, device, sensor, location, app event, Shortcut/App Intent, command, or manual triggers when the platform supports them. A trigger creates an ordinary Run; it does not bypass Workspace Profile, approval, Critical, or Data Send checks.

Desktop monitoring continues only while the app process is running.
Mobile platforms recover overdue work when the app returns to the foreground.
Use Shortcut/App Intent or a trusted event relay when operating-system delivery is required.
Queued events, pause, cancel, and deletion remain explicit so events are not silently discarded.

Skills and Skill Packs

A Skill stores an outcome-focused prompt, typed variables, provider recommendation, allowed tools, required connections and capabilities, maximum risk, and expected Artifacts. Before use, QuickAgent shows the inputs and effective permissions; the preview does not itself grant permission.

Seven built-in Packs provide real Skills for Office, Coding, Smart Home, Research, Browser Automation, Personal Admin, and Incident Response. You can install, update, uninstall, customize, duplicate, import, and export Skills. Imported JSON receives a trust, publisher, signature, permission, risk, platform, and conflict preview. “Keep both” creates an unsigned local derivative rather than inheriting another publisher’s trust.

Artifacts, History, and Undo

Artifacts keep output bytes, metadata, versions, provenance, source Run, and operation manifests. Depending on platform capability, you can open, share, export, or reveal them. Eligible operations expose a reviewed Undo; destructive or uncertain work is never replayed automatically.

History is append-only. Filter it by Safe, Approve, Confirm, or Critical, and export the audit when you need an external record.

Settings

General: appearance, language, Workspace Profiles, shortcuts, Help, and permissions.
AI: providers, endpoints, models, credentials, and live tests.
Tools: catalog readiness, platform support, and setup reasons.
Safety: approval rules, Data Send behavior, device authentication, and scoped Memory policy.
Advanced: recovery and diagnostics intended for troubleshooting.

Keyboard Shortcuts

On macOS, the app maps the displayed Ctrl shortcuts to the platform Command key where appropriate. The in-app Shortcuts page is authoritative and supports customization.

Shortcut	Action
`Ctrl` + `N`	New Chat
`Ctrl` + `L`	Focus the composer
`Ctrl` + `Shift` + `Space`	Pause or resume the active Run
`Ctrl` + `.`	Stop the active Run
`Ctrl` + `1` … `9`	Open Chats through History
`Ctrl` + `,`	Open Settings
`F1`	Open Help & Privacy
`Esc`	Dismiss a dialog or cancel the current non-destructive interaction

Platform Notes

Platform	Important boundaries
macOS	Sandboxed in Debug and Release. File access uses selected roots/security-scoped bookmarks. Screen Recording and Accessibility are separate system permissions. Local process and Shortcut integrations remain capability-tested.
Windows	File, Office/PDF/data, browser, development, remote, and desktop-control features depend on the installed runtime and permissions. Apple-only integrations are unavailable with a visible reason.
iPhone / iPad	Files use the system picker. Camera, microphone, speech, photos, local network, and notifications are requested in context. Background monitoring is not guaranteed; use Shortcuts/App Intents or foreground catch-up.
Android	Files and folders use the Storage Access Framework. Background work follows Android restrictions. Accessibility-based input control is never treated as granted merely because a service is declared.
Web	Uses session-only storage for features that require native persistence and clearly disables unsupported device, file-root, local-process, and background capabilities.

Tips & Tricks

Start with a narrow Workspace Profile and add a tool, connection, or file root only when the Run shows a specific requirement.

Give each request a clear outcome, expected Artifact, and review boundary so progress stays easy to audit.

Use Knowledge for reusable indexed sources and Context for material needed only by the current Run.

When an external effect is uncertain, inspect the destination before retrying; Resume never repeats it automatically.

Troubleshooting

Problem	What to check
The provider cannot run	Open Settings > AI. Check configuration, authentication, model capability, endpoint reachability, timeout, and rate-limit details, then run the connection test.
A tool is disabled	Open Settings > Tools and read the live reason. Review the current platform, permission, connection test, Workspace Profile, and selected root.
A Run looks stuck	Open Runs and the Timeline. It may be paused, awaiting input, approval, Critical confirmation, Data Send review, or budget review. Stop remains available.
A permission changed	Open Help > Overview or Settings, then open system settings. Re-test the affected connection or tool after returning.
An external action is uncertain	Check the target system. Keep the Run paused or Stop it unless you have verified that retrying is appropriate.
An Automation did not fire in the background	Mobile operating systems do not guarantee arbitrary polling. Return to the foreground for catch-up or configure a Shortcut/App Intent or trusted event source.
An Artifact cannot open or export	Check whether its bytes are still available, whether the platform supports the action, and whether the chosen destination permission is still valid.

Privacy, Permissions, and Licenses

QuickAgent stores conversations, Runs, approvals, Knowledge metadata, Skills, and settings locally. Credentials use platform-protected storage. Content leaves the device only through a provider or connected service you configure, and external sends receive a destination-aware preview.

Permissions are requested in context: files/folders, photos, camera, microphone, speech recognition, local network, notifications, screen recording, and accessibility/input control are independent. Removing a permission makes the affected capability unavailable rather than producing a false success.

Open Settings > Help & Privacy for the current platform capability table and Open-source licenses for package and bundled-native notices.

Free and Pro: all agent features remain available in Free. The seven-day trial is ad-free; afterward, mobile may show ads and desktop may show a daily reminder. Pro removes those messages through a yearly subscription or lifetime purchase.