Main Agent
The Main Agent is XainFlow's autonomous AI assistant. Unlike the Flow Assistant inside Flow Studio — a guidance overlay that answers questions while you build workflows — the Main Agent executes tasks end to end: you describe what you need, and it plans, runs, and reports back with the results.
Main Agent is currently available as an Enterprise preview. Capabilities and availability may change before general release.
What It Does
The Main Agent reasons over your workspace, calls XainFlow tools, and manages long-running work as discrete tasks. Each task carries its own brief, plan, conversation history, and outputs.
| Capability | Description |
|---|---|
| Task management | Spawn one-off or recurring tasks, track them in a unified list, archive when done |
| Content generation | Generate images, videos, audio, and full workflows on your behalf |
| External connectors | React to events from Jira, Monday.com, Google Drive, and email |
| Skills | Run reusable recipes for common creative tasks, including motion templates |
| Creative Intelligence | Collect competitor ad references, save them as research, and remix them into original briefs and assets (Creative Intelligence) |
| Scheduling | Run tasks on a cron schedule for repeatable work |
| Brand context | Auto-injects your workspace and project brand assets (logo, colors, voice) into prompts |
| File uploads | Attach documents, spreadsheets, decks, audio, and video to a task — the agent reads and acts on them |
The Agent Workspace
The Main Agent UI is organized around three regions:
- Task list — every active, scheduled, and archived task in one place. Tasks coming from external connectors (Jira, Monday) carry compact badges (e.g.,
BF-2949) and a tooltip with the source item name. - Task detail — timeline-based chat with the agent, organized into thinking groups for multi-step reasoning. Asset references appear as message attachment chips you can preview inline.
- Brief panel — the structured plan the agent is following, including next steps and known blockers.
A recommended skills rail suggests skills relevant to the current task. You can also open the skill picker to invoke any skill manually. The agent home page surfaces Quick Start pills with prepared briefs (product photos, campaign pipeline, brand audit, videos from images, style exploration, platform resizing) so you can start a task in one click.
Uploading Files to a Task
Drag files onto the brief panel — or use the attachment button — to give the agent context. The agent reads the file content (not just the filename) and acts on it.
| Limit | Value |
|---|---|
| Files per message | 10 |
| Max file size | 25 MB |
Supported formats
| Category | Formats |
|---|---|
| Images | PNG, JPG, JPEG, WebP, GIF, SVG, HEIC, BMP |
| Documents | PDF, DOCX, PPTX, XLSX, CSV, MD, TXT |
| Data | JSON, XML, YAML |
| Audio | MP3, WAV, OGG, M4A, AAC, FLAC |
| Video | MP4, MOV, WebM, AVI, MKV |
| Subtitles | SRT, VTT |
| Bundles | ZIP |
You can also attach existing items from your asset library — the asset picker accepts every type the agent can read, not just images.
Spreadsheet workflows are a sweet spot. Hand the agent a CSV of products and ask it to generate a hero image and copy block per row, then save the results back to the library.
Modes
| Mode | Behavior |
|---|---|
| Assist | The agent proposes the next action and waits for your confirmation |
| Auto | The agent executes the plan continuously and asks only when blocked or when high-cost steps are detected |
You can switch modes per task. Auto mode is convenient for trusted recurring work; Assist mode is the default for new tasks.
Brand Context
When the agent generates content, it automatically pulls the active brand logo, colors, and voice from your workspace and project. The hierarchy is:
- Project brand (logo, colors, variables, styles) — wins when set
- Workspace brand — used as the fallback
The same context is exposed to MCP via xainflow_get_context. See Backgrounds & Logos for how to set logos at each scope.
Where to Go Next
- Integrations — connect Jira, Monday.com, Google Drive, and email
- Skills — workspace skills, dynamic skill creation, and motion templates
- Scheduled Tasks — run agent work on a cron schedule
- Models — choose the agent model tier (Balanced I, Balanced II, Smart)
- Creative Intelligence — collect and remix competitor ad references