Productivity
Productivity is the task board — a Kanban board. Every job your agents are doing shows up here as a card, so you can see the whole picture at once.
The tabs at the top of the page switch between four views of the same work:
| Tab | What it shows |
|---|---|
| Kanban | The board: one card per task, in columns from To do to Done. |
| Task Orchestration | How a task's steps connect and where each one stands. |
| Agent Network | Your agents, and the work passing between them. |
| Virtual Office | Your agents at work in a 3D office, with a small Kanban board on the wall. |
The Kanban board #
Work moves across columns from left to right as it goes:
| Column | What it means |
|---|---|
| To do | Lined up and waiting to start. |
| In progress | An agent is working on it right now. |
| Blocked | The work stopped before it passed. The card says why. |
| Review | The work passed C5's final check and waits for you to approve it. |
| Done | Finished. Results are in your Library. |
For work your agents do, Review means the work passed. C5 moves a task there once its final check has passed the work, while Automated Review is off; with it on, work that passes goes straight to Done. Open the task to approve or deny it (see below).
Blocked means the work stopped before it passed. When C5 stops it, the card says why. If C5 needs a decision from you, the card asks for it: write a note, then click Approve or Reject. Otherwise it shows the reason. A long one is cut to two lines: point at it or tap it to read the rest, or open the task to see it in full under Why this is blocked.
You can drag cards between columns to change what happens next. Move something into To do to queue it up, or out of the way if you changed your mind.
A card with AI Enabled off is work for people, and it stays where you put it. Drag one into In progress and it stays there until someone moves it; C5 no longer sends it back to To do after a few idle minutes. A card with AI Enabled on that sits in In progress with nothing running is still moved back to To do, so C5 can start it again.
A sub-task that orchestration created waits in its own block under To do until you approve it. See Sub-tasks.
Creating a task #
Click New Task, describe the work, and click Create Task.
The readiness check #
Before a new task is created, C5 takes a quick look at its title and description to see whether they say enough for the work to start well. The check runs whether AI Enabled is on or off. While it runs, Create Task shows a few moving dots, and pointing at it says Checking task description readiness. It gives up after 90 seconds at most.
If the description looks thin, a box opens under the button: More detail recommended, a short reason, and a list of what seems to be missing. You choose what happens next:
- Edit closes the box so you can add detail. So does Esc, or simply typing in the title or description. Click Create Task again when you are ready, and C5 checks the new text.
- Intent Wizard opens the wizard to sharpen the description with you. It is offered when the wizard is available and you have not used it on this task yet.
- Create Anyway creates the task as it is. C5 does not ask again about the same text.
- Cancel asks first: Cancel task creation? This task has not been created yet. Cancelling now will discard it. Keep goes back; Discard throws the new task away.
The check never stops you creating a task. If it cannot get an answer — the model is busy or offline, or it takes too long — the task is created as if the check had passed. A description the Intent Wizard has already sharpened, and that you have not changed since, is usually not checked again.
The same check runs when you click Auto Create on a task card in Chat, with the same box and the same choices.
Images and files you attach #
When an agent works on a step, it is given the task's attached images, with a few limits so a request is never too big for the model to accept:
- At most five images per step, and about 1.5 MB of images in all. An image sent twice counts once. Large images are scaled down to fit within 1024 × 1024 pixels first.
- Images beyond the limit are not sent, but the agent is told their names, so it knows they are on the task.
- Steps that plan or check the work are told the images' names instead of being sent the images themselves.
- Other files, such as PDFs and documents, are passed on as text where C5 can read them, up to 12,000 characters each, and never as images.
If a model still refuses a request because of the images in it, C5 tries that step once more with the images left out (the agent is told their names), so the work can carry on.
Picking a recipe #
Under Recipe, pick one of your recipes, and the task follows it. Leave it on — No Recipe (Standard Planning) —, and C5 plans the work from scratch. If the recipe you picked has been deleted by the time the work starts, C5 plans from scratch instead.
If C5 knows a recipe that suits what you wrote, it suggests it under the picker, as Suggested recipe: and the recipe's name. Use picks it; Dismiss hides the suggestion.
If the picker is still there when you edit the task later, you can change the recipe and click Save. Once the work has started, the picker is gone.
A recipe you chose on the Create Task card in Chat comes with you: click User Defined on the card, and the task window opens with that recipe already chosen.
If the pipeline hand-off fails #
With AI Enabled and Orchestration both on, Create Task saves the task first, then hands it to the pipeline. If that hand-off fails, you see Task created, but its pipeline hand-off failed. The pipeline will start on its own within a few minutes.
You do not need to do anything. The task is saved. If you created it in To do, C5 starts its pipeline itself within a few minutes, as long as the task is still in To do with AI Enabled on, and has not been archived or deleted. If you created it in another column, its pipeline already started when you created it.
Looking inside a task #
Click any card to open it. Inside you will find:
- The plan — every step, and which ones are finished
- The work — what each agent did and what came back
- The cost — how much this task has used so far
- The files — anything made along the way
- Post-completion recipes — follow-up recipes that start automatically once this task's work is approved
- The history — a full record of what happened, in order
This is the place to look when you want to know why something turned out the way it did.
Opening a task's files #
Once a task has made something, its card lists the files, with Open Directory for its working folder. Anyone who can see a task can open its files and folder this way, not only administrators. Files open on the computer running C5. If one cannot be opened, the card says why:
- That file is no longer on disk, or is not one you can open.
- That file is outside C5's working folder, so it cannot be opened from here.
- Files open on the machine running C5, not from a remote device. — you are using C5 through Remote Control.
A task waiting for review shows Human Review Required inside it. Enter your reasoning, then choose Approve or Deny. The ruling is offered to someone with Edit on Sidebar: HITL Button & Modal in Settings › Access who is also the task's owner or an administrator. A task an agent or C5 itself created has no owner, so anyone with that permission can rule on it. Anyone else sees why instead of the buttons: Needs HITL edit permission to approve or reject tasks., or Only this task's owner or an administrator can approve or reject it. See Human review (HITL).
When you edit a task, the title, description and reasoning boxes each have a small microphone. Click it, say what you mean, and click again to stop — your words land in the box so you can read them over before saving. Reasoning is the one people dictate most: it is the field where you are explaining yourself rather than naming something, and it is quicker said than typed. See Voice.
Sub-tasks #
While it works on a task, orchestration can split part of the task into a sub-task: a separate task on the board for one piece of the work. This only happens with Create Sub-Tasks turned on in Settings › General › Tasks. It is off by default, and while it is off, tasks are never split. See the Tasks panel.
Nothing a sub-task does starts without you. Every new sub-task waits for your approval.
Telling a sub-task apart #
A sub-task's id starts with task_, like many other tasks, so look at its id badge
instead: on a sub-task it is yellow with black text, on its card and in its task window.
Point at the badge and, under the usual Click to copy Task ID line, it shows Parent
task: and the parent's title. Click the title to open the parent task. While you are
editing the sub-task, the parent's name is shown as plain text, so a click cannot throw
your changes away; Save or Cancel brings the link back.
If a link to a task cannot be opened — the task was removed, or it is not one you may see — C5 says That task is not available. It may have been removed, or you may not have access to it., and the window stays on what you had open.
Pending sub-tasks #
New sub-tasks wait in a separate block at the bottom of To do, headed Pending sub-tasks with a count. The block only appears while there are sub-tasks in it, or while you are dragging a sub-task.
Nothing in that block is picked up by orchestration. Agents cannot move a pending sub-task either; only a person dragging it on the board can. Point at the block's heading or the space between its cards (not at a card) to see a reminder of how it works.
Approving a sub-task #
Drag a sub-task up out of the block into To do to approve it. Approving does not start it at once:
- Sub-tasks are created with AI Enabled off. An approved sub-task with AI still off stays in To do for people to work on.
- With AI Enabled on, an approved sub-task is picked up in the normal task order, like any other task waiting in To do.
To let agents do a sub-task, open it, turn AI Enabled on, and click Save. You can do this while it is still pending; it then starts in the normal order once you approve it. On a sub-task that is pending, or approved but not started yet, the Orchestration switch follows AI Enabled and cannot be set on its own. The line under it says where the sub-task stands: On with AI Enabled. Waiting for approval — once moved into the upper To Do area it is picked up by the normal task order., or, once approved, On with AI Enabled. Approved — starts by the normal task order.
You can also drag a pending sub-task straight to another column. Dragged to In progress, it is released for work there: with AI Enabled on, its work starts; with it off, it stays in In progress for people.
Stopping a sub-task #
Drag a sub-task back into the Pending sub-tasks block, from To do or any other column, to stop work on it. It stays stopped until someone drags it out again. A pending sub-task asks for no sign-offs, and dragging one into the block withdraws any it was waiting on; to decide it, drag it out first.
Only sub-tasks can go in the block. Normal tasks cannot be dropped there.
Turning Create Sub-Tasks off stops new sub-tasks from being made. Sub-tasks that already exist stay where they are, pending ones included, until someone moves them.
The To do count #
The count beside To do counts tasks and sub-tasks separately. It shows one number when only one kind is there (0 when the column is empty), and two when both are: 3/2 means three tasks and two sub-tasks. Sub-tasks you approved still count as sub-tasks. Point at the count to see Tasks: 3 and Sub-Tasks: 2, listing only the kinds that are there.
The ⓘ beside the page title explains the same rules under 💡 How Sub-Tasks Are Approved. That is the place to look on a phone, where the block's own reminder does not appear.
Running recipes after a task finishes #
Sometimes finishing a task is only the first step. For example, one task might write a social post and make an image for it. Once that post and image are ready, you might want a recipe that posts to X.com and another recipe that posts to LinkedIn.
Post-Completion Recipes let you line up recipes on a task so they start automatically as soon as the task's deliverables are approved — using the finished work and images the task just made.
When they start #
Post-completion recipes wait until the task is approved and moves to Done:
- If Automated Review is on, the recipes start as soon as C5's review passes the work.
- If the task waits for your sign-off (Human Review Required), the recipes start right after you click Approve.
You can add, remove, or reorder post-completion recipes at any time before the task is Done.
Setting up a recipe list #
When you open a task, the Post-Completion Recipes section shows any recipes already lined up, or No post-completion recipes configured for this task. if none are set yet.
- Open the task and click + Add Recipes (or Manage if recipes are already set, or double-click the Post-Completion Recipes section, or click Edit at the top).
- Under Post-Completion Recipes, open Select a recipe to trigger after completion..., pick a recipe and click Add Recipe.
- Add as many recipes as you need.
- Drag the handle on the left of any recipe — or use its up and down arrows — to put the list in the order you want. Click the trash button to take one off the list.
- Click Save.
Saving and reusing Recipe Bundles #
If there is a set of follow-up recipes you use often — like posting to X.com, then LinkedIn two hours later, then drafting a newsletter — you can save the whole setup as a Recipe Bundle and apply it to future tasks in one click:
- In the Post-Completion Recipes editor, click Recipe Bundles in the top-right corner of the section.
- To save your current list (including its Serial or Parallel mode, custom notes, file filters, and wait timers), type a name in Bundle name and click Save Bundle.
- On any future task, open Recipe Bundles and click Apply next to a saved bundle to load the entire chain right away.
One after another, or all at once #
Choose how the recipes run with the Serial and Parallel buttons at the top of the section. The choice matters once your list has two or more recipes:
- Serial runs them one by one from top to bottom. Recipe 1 starts when your task is approved. Once Recipe 1 finishes and its work is approved, Recipe 2 starts next, and so on down the list.
- Parallel starts every recipe on the list at the same time as soon as your task is approved.
The editor follows the button you choose. Once recipes are on the list, the line under them starts with Serial order: or Parallel fan-out:. While the list is still empty, the example under the picker changes with it: with Serial it describes posting to X.com, then LinkedIn in serial order, and with Parallel, posting to X.com and LinkedIn in parallel order.
Extra instructions, auto-matched file filters, and step delays #
Click Options on any recipe in your list to tailor how that step runs:
- Custom instructions / channel overrides — write a short note for just that recipe. For example, tell your X.com recipe "Keep it punchy and add 3 hashtags" and your LinkedIn recipe "Use a warm, professional tone for founders."
- Deliverables to pass to this recipe (with Auto-match) — by default (All types), C5 hands every approved piece of work and image from the task to the recipe. When you add a recipe whose purpose is clear (such as a social graphic post, a code deployment, or a design review), C5 automatically pre-selects the matching file types (Document, Code, Design, or Image) and marks them Auto-matched. You can click any filter pill to change the selection, or click Auto-match to restore C5's suggested file types.
- Step Delay Offset (Wait Before Trigger) — if you want a pause before a recipe starts (for example, waiting +2 hours after an X.com post before publishing on LinkedIn, or +24 hours before sending a follow-up summary), pick a delay from the dropdown (Immediate, +15 minutes, +30 minutes, +1 hour, +2 hours, +4 hours, +12 hours, or +24 hours).
Watching the chain and jumping between tasks #
On the board, any card that has post-completion recipes — or was started by one — wears a small Chain badge:
- Chain 2 (indigo) — 2 recipes are lined up and waiting for this task to finish.
- Chain 1/2 (amber) — step 1 of 2 is running now.
- Chain 2/2 ✓ (green) — every recipe in the chain has finished.
- Chain 1/2 ! (red) — the chain stopped and needs your attention.
Inside the task, the Post-Completion Recipes section shows which step is running and gives you one-click buttons — Jump to Triggered Task, Parent Task, and Root Task — so you can hop straight from the original task to the follow-up tasks it created, and back again.
When a recipe step is waiting on a Step Delay timer, the task shows a Chain scheduled badge along with the time the next recipe is set to start. If you do not want to wait, click Run Now to skip the timer and start that recipe immediately.
If a chain stops #
If a recipe in the chain cannot start (for example, because that recipe was deleted) or a step in a serial chain fails, the chain stops and tells you why in plain words. Once you fix the problem, open the task and click Resume Chain to start the stopped step and keep the rest of the chain going.
Seeing your agents work together #
The Agent Network view draws your agents as a picture. Lines connect agents that are passing work to each other, and they light up as work moves.
It is a fast way to spot a bottleneck — if everything is pointing at one busy agent, you will see it right away.
Keeping things organized #
As you use C5 more, the board fills up. A few ways to keep it under control:
- Filter by status, agent, or date using the controls at the top.
- Tag related work so you can pull it up together later. See Tags.
- Group by project so client work stays separate from your own.
- Archive finished work you no longer need to see.
Changing what happens next #
You are not stuck with the plan C5 made. You can:
- Reorder the queue by dragging cards
- Pause a task and come back to it
- Cancel a task you no longer want
- Retry a task that failed, after you fix what caused it
The LLM Queue badge #
The small badge in the page header, beside the title, is the LLM Queue: how many agents are running right now against how many may run at once, for example 2/3. Its dot is the queue's health — green while agents are working, grey while nothing is queued, yellow while a step is waiting to retry, red when a step has been stuck for longer than your stalled-job timeout, and grey again while Master Pause is on.
When a paid provider is first in Task Priority, the badge shows one pair per kind of provider, each behind an icon: a server for local models and gateways (OpenClaw, Hermes Agent, PicoClaw, LM Studio, Ollama), a cloud for paid providers. So ⛁ 1/1 · ☁ 0/2 means one local agent running of one allowed, and none of two cloud agents. The two allowances are the Concurrent Agents and Cloud Concurrent Agents sliders in Settings › General — see Settings.
Hover the badge for the detail: a row per kind of provider named after its slider, and Active Agent Slots, one line per running agent with the provider serving it, the server or cloud icon beside it, its role, and the step it is on. A gateway always shows under the server icon, because that is the allowance it counts against — a gateway may still forward a request to a paid model. Administrators also get a restart button here for stalled steps.
Pausing everything #
Master Pause, the pause button in the page header, halts all new AI task steps, for everyone using this C5. Its tooltip says Halt all new AI task steps. Click to pause., and while it is on the button glows red: AI orchestration is halted. Click to resume. Click it again to resume.
It is for administrators, and for anyone given Edit on Orchestration Master Pause Button in Settings › Access, which is enough on its own. For anyone else it is greyed out, with Needs the Orchestration Master Pause permission (Settings › Access). in its tooltip. Permission to change other settings does not include it. If C5 refuses the change, the button goes back to how it was.
When something fails #
Failed tasks turn red and move to Blocked. Open the card and C5 tells you what went wrong in plain words.
Most failures are simple: a missing API key, a file that moved, a spending limit reached. Fix the cause and click Retry.
If a task keeps failing, see Common issues.
