Complete user guide
This guide walks through real usage: install, sign in, configure BYOK, open a trusted workspace, use Plan and Act, approve tools, review diffs, run checks, manage settings, and troubleshoot common issues.
Step 1
Use this when you want the fully branded desktop app with DevCoreAI opened by default.
Use this when your team already works in VS Code, VSCodium, Cursor, or another compatible editor.
Step 2
Step 3
BYOK means bring your own key. Your model provider bills inference usage directly. DevCoreAI provides the IDE, agent workflow, settings, approvals, dashboards, platform limits, and team experience.
| Setting | What it does | When to change it |
|---|---|---|
| Provider | Selects which AI provider DevCoreAI calls for the current workspace. | Choose the provider where you already have an API key, billing, and model access. |
| API key | Authenticates requests directly from your local DevCoreAI client to the provider. | Paste only provider keys you control. Rotate the key if it was exposed or shared. |
| Model | Controls quality, speed, context length, and provider cost. | Use stronger models for architecture, security, refactors, and multi-file work. Use faster models for quick edits. |
| Reasoning effort | Controls how much planning depth the model uses when supported. | Use higher effort for complex tasks and lower effort for quick, low-risk work. |
| Plan/Act split | Allows one model for planning and another for implementation. | Use a strong planning model with a cost-efficient implementation model when your workflow needs both depth and speed. |
Step 4
Use File > Open Folder so DevCoreAI can detect workspace roots, rules, git state, package files, and test commands.
Terminal execution is blocked in untrusted folders by design. Trust a workspace only when you trust its scripts and dependencies.
Add .devcoreai/global-rules.md and .devcoreai/agent-rules.md for durable standards. Legacy rule files remain compatibility fallbacks.
Use .devcoreaiignore for secrets, generated output, archives, dependencies, build artifacts, and private customer data.
Step 5
Ask DevCoreAI to analyze the codebase first. Good prompts include the goal, constraints, affected area, and what must not break.
Check that DevCoreAI found existing patterns, reusable files, risk areas, test commands, and a minimal implementation path.
Read file edits, commands, browser actions, and MCP calls before approval. Avoid auto-approving destructive commands, secret changes, billing, auth, migrations, or releases.
Use inline diffs, editor tabs, and checkpoint review to verify that changes match the plan and do not rewrite unrelated code.
Let DevCoreAI run the detected lint, typecheck, test, and build commands. If checks fail, keep the task open until the agent fixes its own changes.
Ask for a final report that lists changed files, reused patterns, checks run, failures, and any manual follow-up.
Reference
Shows sign-in state, user identity, plan, dashboard links, and sign-out actions. Use it after browser login to confirm the IDE stored the token correctly.
Controls BYOK provider, API key, model, reasoning effort, and advanced Plan/Act model separation. Use this before any task that requires model calls.
Shows the active folder, workspace roots, trust state, rules, ignore files, and effective policy. Use it when DevCoreAI says no folder is open or cannot read expected files.
Controls whether terminal execution is allowed, which commands require approval, and how untrusted workspaces are handled. Keep destructive commands manual.
Controls browser automation for local web apps, auth flows, screenshots, and page verification. Enable when testing UI or dashboards.
Manages MCP servers, tools, hooks, workflows, and plugin-like extensions. Enable only trusted tools because they can expose external systems.
Controls telemetry, local-first behavior, sensitive context handling, and reporting. Organization policy may lock some values.
Controls preview features, automation helpers, inline diffs, checkpoints, and advanced UI behaviors when available.
Advanced
Approve test, build, install, and diagnostic commands when they are expected. Keep deletion, deploy, migration, and credential commands manual.
Connect external systems only when the server is trusted, scoped, and necessary. Review tool permissions before enabling.
Use checkpoints before large refactors, dependency upgrades, migrations, or generated multi-file changes.
Troubleshooting
Trust the workspace only if you trust the project. Otherwise keep terminal execution disabled and approve file reads only.
Reload the IDE window, open Account settings, and confirm the extension stored the returned auth token. If it persists, sign out and repeat the browser login flow.
Use File > Open Folder instead of opening a single file or archive. Confirm the Workspace settings page shows the same primary folder.
Check API key, provider billing, model access, selected model name, proxy settings, and whether the provider allows your region or organization.
Use the Download page selector and verify the file name, platform, architecture, and checksum before installing.
For production code, start with analysis, approve a plan, then let Act mode implement with reviewable diffs and checks.
Learn deep planning