CoHunter Guide

CoHunter Guide

CoHunter lets you build a team of autonomous hunters that work for you. Each hunter has its own job, its own memory, and its own budget. They run in the background, work on several targets at once, and pass findings to each other.

This guide shows you how to build one, run it, control it, and make it better over time. For an overview of what CoHunter is, see the CoHunter feature page.

Authorized targets only. Point hunters only at systems you own or have written permission to test. Hunters follow your permission modes and budgets. See Permission Modes.

Two ways to build a hunter

You can build a hunter yourself, or ask the AI to build the team for you. Most people do both: let the AI set up the crew, then step in and improve the hunters that matter.

• Build it yourself. Open the CoHunter panel and create the hunter by hand. You choose the rules and the standards.
• Ask the AI. Say something like "Build me a recon hunter and a reporter and run them." The AI drafts the hunters, assigns the work, and brings back findings.

Open CoHunter

• Desktop: click the team icon.
• Web: open the Tools menu and choose CoHunter.

Your list of hunters is on the left. The work area is on the right. Click Back to main chat at any time. Your hunters keep working while you are away.

You can also just say: "Show me my hunters."

1. Create a hunter

Click + New and fill in these fields:

Field	What to enter
Name	A short name that describes the job, such as ScopeRecon, ExploitOperator, or VerifyReporter.
Instructions	The most important field. Describe the hunter's role and rules, and how you want it to work.
Model	The fast default model, or a heavier model for difficult logic.
Tools	All tools, or a limited set that you choose.
Budget	The maximum number of tool calls for one task (for example, 120), or Unlimited.

Write clear instructions. The clearer they are, the better the hunter works. For example:

You own recon. Map the attack surface, enumerate carefully, flag promising leads, and hand them to the team. You never exploit anything. You stay strictly in scope, and if something is out of scope you stop.

To build a whole team with the AI, you can say:

Create three hunters. ScopeRecon maps the target. ExploitOperator proves bugs with safe proofs of concept. VerifyReporter writes verified findings to submission standard. Give each a budget of 150 calls.

2. Start a hunter

1. Pick a hunter.
2. Type its mission, for example "Enumerate the API surface and post anything interesting to the team board."
3. Click Run.

You can also say: "Send ScopeRecon at company.example.com and message me when it finds something."

Each hunter works in the same folder and on the same target as the chat you started it from, so there is nothing to set up.

3. Let the team coordinate

Hunters work together without you managing every handoff:

• A hunter that finds something passes it to the teammate built to finish it. Recon passes to exploit, and exploit passes to the reporter.
• Hunters post to a shared team board, so nobody repeats work that is already done.
• Hunters message each other during a task and ping you only when something deserves your attention.

To get this, build hunters with complementary roles, start the first one, and let the work flow.

You can also say: "Run the whole team on this target. Do recon, test the best leads, and write up anything verified. Only wake me for confirmed findings."

4. Watch a hunter work

Open any hunter to see a live feed of every command and result as it happens. The header shows:

• the status (running, paused, stopped)
• the budget used, for example 59 of 200 calls
• a context meter that shows how much room the run has left. The hunter manages this itself, and you can compact it manually if you want.

You can close the view and open it again in the middle of a run. The full history is always there.

5. Stay in control

Every hunter has instant controls. You can also ask the AI to use any of them for you.

Control	What it does
Pause / Resume	Freezes a run and lets you continue it later.
Stop	Ends the task cleanly.
Clear	Wipes the session so you can start fresh.
Compact	Shortens a long run so it can keep going.
Edit	Changes the hunter's instructions, model, tools, or budget at any time.
Budget	Raise it, lower it, or set it to Unlimited while it runs. If a hunter hit its limit, raise the budget and it continues where it stopped.

You can also say: "Pause ExploitOperator, raise its budget to 300 calls, and resume it."

6. Memory

Every hunter has its own learning memory that carries across sessions and across targets. This is what makes a hunter improve over time instead of starting over.

• View and edit it. Open a hunter and choose Memory. You can see what it has learned about your preferences, the mistakes it corrected, and the approaches that worked. You can add notes, edit them, or clear them.
• It learns from your corrections. Correct a hunter once and it remembers the lesson.
• It keeps targets separate. Information that belongs to one target stays with that target and does not appear on the next one. What it learns about you is permanent. You can promote a target note into permanent memory when it is worth keeping.

You can also say: "Remember: I always want proof of impact before any handoff." The hunter saves it and follows it from then on.

7. Run many hunters at once

Hunters run in parallel and are fully isolated from each other. The same hunter on two targets runs as two separate jobs that never mix. Start a team across every in-scope asset and let them work at the same time.

You can say: "Run ScopeRecon on all three assets at once and give me a summary on the board when they are done."

8. Use hunters in workflows

A hunter can be a step in an automated methodology. Add a CoHunter to a workflow, give it its instructions, and let its results feed the next steps. This combines your repeatable process with your team of hunters. See the Workflow Engine.

Tips for better results

• Give each hunter one clear job. A specialist hands off better than a hunter that tries to do everything.
• Start with a small budget. Watch how the hunter spends it, then raise the budget for hunters that earn your trust.
• Let them talk to each other. Build recon, exploit, and report hunters and the work routes itself.
• Use the AI for setup and your judgment for decisions. Let the assistant build the crew, then step in to redirect, pause, and decide what is real.
• Teach the memory on purpose. A few clear preferences, such as "prove impact," "stay in scope," and "report in this format," improve every future run.

Related

• Sub-Agents and Swarm
• Background Tasks
• Project Memory
• Permission Modes
• Web Console