Skip to main content
A task is how you put Neo to work. You describe what you want (a target, an objective, a question) and Neo plans, executes, and returns results. Every pentest, code audit, triage run, or ad-hoc investigation is a task. Tasks accumulate in your workspace and feed into projects, issues, and reports over time. Every task runs in one of three modes. The mode you choose controls how much you steer and how much Neo decides on its own.

​
Fast mode

Neo skips the planning step and dispatches agents immediately. Best for focused, well-scoped asks: targeted scans, quick follow-ups, retesting a fix, or any task where the objective is narrow and clear. Faster. Lower cost.

​
Thorough mode

Before executing, Neo asks clarifying questions, builds a structured plan, and spins up parallel sub-agents for each part of the plan. This is the mode for depth: full-scope pentests, broad reconnaissance, multi-phase assessments, and any open-ended objective where a fast pass would miss things.
1

Clarify

Neo asks follow-up questions to narrow scope and understand constraints. The better your answers, the more targeted the plan.
2

Plan

Neo produces a structured plan using your answers, project context, and any prior task results. It reads files and searches knowledge during this phase but does not run active tools until the plan is ready. Neo pauses here and waits for your go-ahead. If you want something changed, say so: Neo revises the plan and comes back for approval before it starts.
3

Execute in parallel

Neo spins up parallel sub-agents for independent parts of the plan and runs them simultaneously, which is what makes it effective for broad objectives.
4

Verify

Before surfacing any finding, Neo independently re-tests it to confirm exploitability and severity. Findings that cannot be verified are discarded, not reported.
Long running. Higher cost. Better output quality.

​
Goal mode

Describe the outcome you want instead of the steps to get there. Neo plans the work, runs it, checks its own results, and comes back to you only when it needs a decision it can’t make on its own. Goal mode builds on Thorough mode with one difference in intent: the job is not finished until the outcome is reached. If one approach doesn’t get there, Neo tries another, and it can keep working for much longer than a Thorough task to get you to the goal you set. That persistence is what makes it the most expensive mode. Because Goal mode can run for a long time, pairing it with a lower-cost model is usually the most economical way to reach the outcome. Neo’s persistence does the heavy lifting; the model doesn’t need to be the most capable one available. Longest running. Highest cost. Built to reach the outcome.

​
Choosing the right mode

Default to Thorough for anything open-ended or broad. Use Fast when you have a specific, well-scoped ask and want results quickly. Use Goal when you can state the outcome clearly and want Neo to get there on its own, checking in only for real decisions.

​
Advanced execution

Modes determine how Neo approaches a task. Advanced execution provides controls for refining the run, including memory, tool review, verification depth, and execution budget. It is available from the settings icon in the chat window, and the settings you choose apply to the task you are about to start. Use past memory On by default. Turn it off to run the task without recalling context from previous tasks, which is useful when you want Neo to approach a target fresh. Auto tool review On by default. Before Neo runs a command, script, HTTP request, or file write, a separate reviewer checks the call against Neo’s data-egress policy and blocks it if it would:
  • Exfiltrate secrets: send Neo’s own credentials, tokens, or environment variables to a destination outside the task’s approved set
  • Delete sandbox data: destroy or disable the sandbox workspace, task directories, or execution records the task depends on
  • Upload your files to an external site: move sandbox files or your data to a host you did not specifically name, such as a public file-sharing or paste site
Approved destinations are your connected integrations, the in-scope target, and any host your prompt specifically names as the place to send data. A vague instruction like “save this somewhere” does not approve whatever site Neo picks. Normal testing is not affected: authenticating to a service with its own credentials, sending payloads to the target, and out-of-band callbacks that confirm a finding all run as usual. A blocked call does not run, and the task shows which rule it broke. Turn auto tool review off only when you want tool calls to run without review. Verification level Sets how many execution and verification cycles Neo runs before it reports. Each cycle re-tests findings and re-runs execution where the verifier asks for it. The default follows the mode you selected. You can turn verification off entirely, or choose Standard (one cycle), High (three), or Maximum (five). More cycles means more independent confirmation of each finding, and a longer, costlier task. Execution budget Sets the maximum number of steps a task can take before it stops. The default range is 45 to 60 steps. Focused (up to 25 steps) suits tightly scoped checks, while Extended (up to 100) and Maximum (up to 200) give Neo more room when a task needs to explore.

​
What a task produces

When a task completes, you get:
  • Findings: confirmed vulnerabilities with full evidence, automatically filed to the Issues dashboard if integrations are connected
  • Trace: every command, decision, and reasoning step, available for audit or replay
  • Report: a summary of what was tested, what was found, and recommended next steps
  • Files: any artifacts produced during the run, including scripts, payloads, and screenshots
Tasks stay open after completion. You can ask follow-up questions, dig deeper into any finding, or chain directly into a new task from where the previous one left off.