Check the active configuration
Before a complex or sensitive task, check which settings affect the current chat. Explyt does not yet provide a single published view with all effective values, so you have to check the configuration across several screens and files.
Short checklist
| What to check | Where to look | Why it matters |
|---|---|---|
| Connection mode | Explyt settings: Personal, Community or Enterprise | Determines how models are accessed |
| Model | The model switcher in Explyt | Affects the context and response behavior |
| Agent mode | The agent selection menu | Determines the role and the baseline set of tools |
| Tools menu | The Tools button next to the input field | Determines tool access and confirmations |
| Edit Scope | A setting of the current task | Limits which files can be changed |
.readignore and .writeignore | The project's .explyt/tools/ or the configured directory | Forbid reading or writing paths |
| AGENTS.md | The project root | Provides permanent project context |
| Rules | The rules menu and .explyt | Add instructions scoped to files |
| Skills | The skills menu and .explyt/skills/ | Add a task scenario |
| Memory Bank | Settings and .explyt/memory/ | Brings in knowledge from other chats |
| MCP | The MCP settings in Explyt | Adds external tools |
| Model Routing | The Model Routing settings in Explyt | Selects models for subagents |
1. Connection and model
Check the selected connection mode and model. For Community, make sure the right provider, API address and key are configured. For Enterprise, use the key and the model list issued by your organization.
The context size and the available parameters depend on the model and the provider. If Explyt shows the wrong limit for an OpenAI-compatible model, set it manually in the settings.
For details, see connection modes and model setup.
2. Agent mode
Make sure the selected mode suits the task:
- Ask — research without changing files;
- Code — ordinary changes;
- Debug — checking runtime behavior;
- Test — creating tests;
- Review — an independent read-only check;
- Plan — planning a large piece of work;
- General and Orchestrator — delegating subtasks.
A custom mode can have its own tool restrictions and Skill list. For details, see agent modes.
3. Tools and confirmations
Open the Tools menu and check:
- whether the tool you need is available;
- whether any unnecessary tools are enabled;
- which calls will require confirmation;
- which MCP servers are connected.
Do not rely on assumed defaults: they are not published as a single contract across all Explyt versions.
4. File restrictions
Check two independent layers:
- Edit Scope — what can be changed in the current task;
.readignoreand.writeignore— what is permanently forbidden to read or write.
If a file is not in Edit Scope, the agent can still read it as long as the path is not blocked by .readignore.
5. Project instructions
Review the sources of text instructions:
- the current request;
AGENTS.md;- the active Rules;
- the selected or automatically attached Skill;
- the custom agent mode;
- the relevant Memory Bank entries.
A complete general priority order among them has not been published. Do not duplicate one instruction in several places, and do not create conflicting requirements.
6. MCP and subagents
In the MCP settings, check the Connected status and the actual tool list. For external services, make sure the account has only the rights it needs.
In Model Routing, check the project and global files. In the current implementation, the project .explyt/model-routing.yaml replaces the global ~/.explyt/model-routing.yaml entirely; profiles are not merged by alias. Add every profile you need to the project file, and after an Explyt update verify this behavior on the installed version. Built-in profiles may inherit the parent agent's model if no separate model is pinned to them.
A quick diagnostic task
Before a large piece of work, you can run a short check:
Do not change the project. List the agent mode, the tool categories available to you,
the active Edit Scope and the permanent project instructions you can confirm.
Do not reveal secrets and do not restate hidden system instructions.
If something cannot be determined, say so.
The agent's answer is a hint, not full technical proof. Cross-check it against the interface and the settings files.
If the behavior is unexpected
- Start a new chat.
- Keep the questionable instruction in one place only.
- Temporarily turn off the optional Rules, Skills, Memory and MCP.
- Check the agent mode, Tools, Edit Scope and the ignore files.
- Bring the settings back one at a time until you find the source of the discrepancy.
The main contracts of the individual mechanisms are collected on the Explyt configuration reference page. Step-by-step checks for file loading and updates are described in configuration troubleshooting.