Work agent
The surface for multi-step assignments. In the ZeroTwo workspace screenshot, Work sits beside Chat and Code, with a task list, a Library and Schedules in the sidebar.
Plus and above. Not included on Free.
AI agent platform
An AI agent platform for work you can supervise and review
Give ZeroTwo a defined outcome, authorized sources and clear review boundaries. Follow the assignment, inspect the files it produces, and decide what moves forward. Start with one bounded task so you can judge it against a finish line.
Work agent runs need Plus ($14.99/mo) or above. The Free plan does not include the Work agent.
Decide first
Decide whether an agent fits this task
An agent run earns its overhead only when the outcome is clear, the path is not scripted and a person reviews the result. Otherwise a simpler tool is the better call.
A one-off, goal-led assignment that spans several sources
An agent run
A research brief, an analysis or a draft pack: the outcome is clear, the path is not scripted, and a person reviews the result.
The same process, every time, on a trigger or schedule
Repeatable workflow automation
When the steps never change, orchestration beats re-delegating each run.
A question, a quick draft or a back-and-forth
No plan or tools are needed, so skip the overhead.
Zero tolerance for variance: payments, filings, irreversible actions
Deterministic automation or a person
Models are probabilistic. Use ZeroTwo to prepare and check the work, not to be the last line of defence.
Where your files and context live is a separate question from how a run executes. To organize project context before you delegate, start in the workspace; this page covers the run itself.
One assignment, start to finish
What happens during a run
Five stages, and who owns each. Tags mark what is a product feature and what is an instruction you give or a check you perform, so nothing here reads as a promise the product does not make.
Write the outcome and the boundary
Your instructionWhat you set
One sentence naming the deliverable, the finish line and what is out of scope.
Check before moving on
Could a colleague read the brief and tell when the work is done? If not, tighten it before any run.
Attach only authorized context
Your instructionWhat you set
The files, links and connected tools the assignment may use. Strip fields the task does not need.
Check before moving on
Does every source belong to you or your client to use? Anything sensitive that the output does not need stays out.
Ask for the plan before the work
Your instructionWhat you set
Request a step list first, and name which steps need your sign-off (external sends, calculations, client-facing text).
Check before moving on
Does the plan match your brief? Edit the brief, not the output, when the plan is wrong.
Run with the models and tools on your plan
Product featureWhat you set
Work agent runs use the models, files, web search and connectors your plan includes (see the access table below).
Check before moving on
Is each finding tied to a source you can open? Restate the brief if the run drifts from the outcome.
Inspect the artifact, then approve
Your reviewWhat you set
Open the finished file. Verify numbers, quotes, links and exports yourself before anything leaves your hands.
Check before moving on
Treat the output as a draft until a qualified person approves it. Record what you had to fix.
Tools and environments
What tools and environments can a run use?
What a run can reach is set by your plan and by what you authorize. The list below is limited to what the pricing catalog and the workspace screenshot show.
ZeroCode
The code agent, for assignments whose deliverable is a code change.
Plus and above.
Deep research
A longer research mode for source-heavy briefs. Pair it with a brief that names which sources count.
Limited on Free; 150 a month on Plus (uses credits); included on Pro and above.
Connectors and plugins
The plugin menu in the screenshot lists Chrome, Google Drive, GitHub, Firecrawl, Browser, Computer Use, Cloud Computer and Sites. Availability, authorization and plan limits vary, so read the integrations page before designing a process around one.
available integrationsZeroTwo connectors
Authorize only what the assignment needs.
Web search
Serper search on Free, Exa on paid plans. Firecrawl credits (1,500) start at Pro.
Per the plan table below.
Model choice
Free has limited model selection, Plus unlimited fast models, Pro a better selection and Ultra all models. The composer in the screenshot shows an Auto selector.
Access by plan
Which plan runs which agent
Prices and allowances come from the same catalog as the pricing page. Pro 2x and Business exist too; the plans and usage limits page has the full list.
| Capability | Free$0/mo | Plus$14.99/mo | Pro$29.99/mo | Ultra$120/mo |
|---|---|---|---|---|
| Work agent | No | Yes | Yes | Yes |
| ZeroCode (code agent) | No | Yes | Yes | Yes |
| Deep research | Limited | 150 / month | Yes | Yes |
| Web search | Serper | Exa | Exa | Exa |
| Firecrawl credits | None | None | 1,500 | 1,500 |
| Monthly credits | None | 1,650 | 3,500 | 14,000 |
| File uploads per task | 1 | Multiple | Multiple | Multiple |
| Model selection | Limited | Unlimited fast models | Better selection | All models |
Source: ZeroTwo's pricing catalog in this repository. Plus is highlighted because it is the first plan on the pricing page with Work agent access. Allowances and features can change; the pricing page is the source of truth.
Supervision
Where you stay in control, and where you should not
ZeroTwo organizes the work; it does not make weak evidence true or remove the need for human judgment.
Controls you set
- The scope of the brief and the finish line.
- Which files and connectors the run may use.
- Which steps need a person's sign-off.
- Whether a draft ever leaves the workspace.
This page does not claim that every external action is gated by the product. Treat approval as your policy and keep consequential actions with a person.
What happens when a step fails?
This page does not document what state a failed step retains, so plan the recovery yourself and test it in your pilot.
- Missing context
- The brief or files do not contain what a step needs. Add the file or narrow the question; do not let the run guess a number you could supply.
- A tool or connector is unavailable
- The integration is not on your plan, is not authorized, or fails mid-run. Re-authorize or finish that step by hand, then continue from the saved brief and files.
- An export or file step fails
- Ask for the content in a plain format first, check it, and then convert. Keep the source files so the step can be repeated.
- The run drifts from the outcome
- Restate the finish line and the review points, and start a fresh bounded task if the thread has become cluttered.
Agents are probabilistic and tool access is permissioned. Never let an unreviewed run make an irreversible, regulated, financial or customer-facing decision.
Pilot
Run one pilot and measure your own cost
We do not publish a per-run price, because it depends on the task, the model and the tools. A single pilot gives you a number you can trust.
Pick a task you could already judge: one you have done by hand, with a known good answer.
Write the brief with the outcome, authorized sources, and the steps that need your approval.
Run it once on your plan. Do not coach it mid-run; note where you wanted to.
Review the output against the checks above and list every fix you made.
Record credits used against your plan allowance, the rework time, and any failed step. That is your cost example.
- Plus: first listed plan with Work agent, 1,650 monthly credits
- $14.99/mo
- Pro: adds Firecrawl credits and a larger allowance, 3,500 monthly credits
- $29.99/mo
- 100 free credits to start plus 20 bonus credits every day you log in, limited tools, no Work agent. Fine for judging chat, not a Work run.
- Free
Questions before you delegate
Can I stop or revise a running assignment?
You can always revise the brief and run again. This page does not document each pause or stop control, so confirm how to interrupt a run during your first bounded task before relying on it for anything long.
Which actions require approval?
Approval is a policy you set in the brief, not a promise that every external action is gated for you. Name the steps that need your sign-off, authorize only the connectors the task needs, and keep irreversible, regulated, financial and customer-facing actions behind a person.
What survives when a tool fails?
This page does not document what state a failed step retains. Keep your brief and source files in the project so any step can be repeated, and test it on purpose in the pilot by withholding a file or an authorization and noting what the run keeps.
What can the agent access in my account?
The files you attach, the web search on your plan, and the connectors you authorize. Features differ by plan: Work agent starts at Plus; deep research is limited on Free and has a 150-a-month allowance on Plus. Anything you do not attach or authorize is out of scope for the brief.
How are task usage and retries charged?
Paid plans include a monthly credit allowance: 1,650 on Plus, 3,500 on Pro and 14,000 on Ultra. We have not published a per-run or per-retry price list, so measure it: run a pilot and record the credits it uses. See the pricing page for current plans.