Code Agents / ACP
Overview
This page explains how Ghast AI's more advanced local coding and automation capability should be understood, and when ordinary users actually need to move into this path.
How to interpret this capability
For most users, Code Agents / ACP do not need to be understood through protocol details.
A more practical way to read them is this: when you no longer want Ghast AI only to answer inside chat, but want it to keep working on deeper coding or automation tasks inside a local workspace, you are entering this advanced path.
These capabilities currently depend on Companion starting local code-agent sessions on the machine to carry heavier local tasks.
What ACP means in practice
For ordinary users, ACP can be treated as the access and runtime path for these local code-agent sessions.
You usually do not need to learn the mechanism by itself. The two important points are:
- it supports deeper local coding and automation tasks
- it is not a layer that needs separate attention during first-time setup
The practical question is not the protocol. The practical question is whether you have actually reached a stage where local code agents are needed.
When this path becomes important
Code Agents / ACP become more relevant when:
- you want Ghast AI to participate in deeper local coding work
- you are preparing to use Ghast AI around a workspace for ongoing work
- you need heavier and more persistent local tasks than ordinary commands
- the base setup is already complete and you are moving into advanced local coding or automation use
If you are still focused on everyday chat, page understanding, basic workspace use, or ordinary command tasks, this page is usually not the priority.
When you do not need it yet
A steadier rule is simple: if the base path is still not working smoothly, you do not need to prioritize Code Agents / ACP yet.
That includes cases where:
- Companion is not being recognized reliably
- the workspace is not connected yet
- the command runtime and other basic local capabilities are not stable yet
- your current task still belongs to ordinary Q and A, page understanding, or light local work
It is usually better to get the base capabilities working first, then move into this heavier local-agent path.
What to confirm first
Before using this capability, confirm the following:
- Companion is installed and recognized by the extension.
- The issue is not actually installation approval, auto-pairing, or another base connection problem.
- A defined workspace is connected and
workspacemode remains the default starting point. - The command runtime and other basic local capabilities are already working.
- Your current task genuinely needs deeper local coding or automation rather than ordinary chat.
This order keeps the connection path, boundary, and workspace setup clear before entering a heavier local-agent stage.
How to think about it in daily use
For most users, this capability is best understood as:
- not a basic onboarding entry point, but a deeper path
- not a replacement for all ordinary conversation, but a way to handle heavier local coding work
- not an unrestricted default capability, but a layer built on top of Companion, workspaces, and the local boundary
That is why Code Agents / ACP should be used when there is a clear need for them, not just to make every capability feel turned on.
If it does not work as expected
If you have entered this stage but local code-agent capability is still unavailable, use this order:
- Confirm that Companion is connected normally.
- Confirm that the workspace, permission mode, and command runtime are already working as expected.
- If the issue remains, continue with Diagnostics and Self-Check.
- If the problem looks more like MCP, installation, recognition, or advanced-environment readiness, continue through the related troubleshooting path.
This matters because many advanced-capability failures still begin earlier in the installation, connection, workspace, or base local-capability path.
Code Agents / ACP are part of Ghast AI's deeper local coding and automation path. For most users, they should be understood as an advanced layer that sits on top of Companion, workspaces, and base local capabilities, and should be used only after the base setup is already in place.
