The Inline Assistant brings Oppla’s AI directly into your code editor so you can get context-aware help without leaving the file. Use it for quick explanations, suggested edits, one-line fixes, documentation generation, and other small-scale tasks that benefit from immediate, low-latency assistance. This page is a concise, actionable stub that explains core concepts, usage patterns, configuration options, privacy considerations, and troubleshooting. Full UX screenshots and GIFs will be added later.

​
What is the Inline Assistant?

  • Lightweight, in-place AI that analyzes the current buffer and surrounding context.
  • Designed for short interactions: explain this function, suggest a fix, write a docstring, or convert a code snippet.
  • Works with local or cloud models depending on your AI Configuration and AI Rules.
  • Complements Edit Prediction (predicts edits as you type) and the Agent Panel (for multi-file or higher-risk tasks).

​
Core capabilities

  • Explain code: produce concise, line-by-line or function-level explanations.
  • Quick fixes: propose single-file, low-risk edits (format, small refactors).
  • Documentation: generate docstrings, README snippets, or code comments.
  • Examples: produce example usage for functions or APIs.
  • Small refactor hints: suggest variable renames or extract helpers (applied only with approval).
  • Test suggestions: propose simple unit test templates for the current function.

​
Quick start

  1. Open a code file in Oppla.
  2. Place the cursor on a symbol, select a code range, or highlight a function.
  3. Invoke Inline Assistant:
    • Command Palette: AI: Inline Assist
    • Default keybinding: Cmd-Shift-A (macOS) / Ctrl-Shift-A (Linux)
  4. Choose the action (Explain, Suggest Fix, Generate Tests, Add Docstring).
  5. Inline suggestions will appear as ephemeral edits or as a preview panel — review before accepting.
Tip: Inline Assistant prefers small scopes (a single function or a code block) for the best results and lowest latency.

​
Keyboard & UX

  • Open Inline Assistant: Command Palette → AI: Inline Assist
  • Accept inline suggestion: Tab
  • Apply suggestion as patch: Cmd-Enter / Ctrl-Enter
  • Dismiss: Esc
  • Cycle alternatives: Cmd-Right / Cmd-Left (configurable)
  • Trigger contextual help hover: Alt-Enter (or platform equivalent)
Customize these in your keymap. See Key Bindings for examples and advanced configuration.

​
Configuration & settings

Inline Assistant settings live in the AI section of your settings file or preferences UI. Example settings (user-level): 
{
  "ai": {
    "inline_assistant": {
      "enabled": true,
      "local_first": true,
      "max_context_lines": 200,
      "suggestion_confidence_threshold": 0.6,
      "auto_accept_trivial_fixes": false
    }
  }
}
  • local_first: Prefer a configured local runtime for privacy and lower latency.
  • max_context_lines: How much surrounding code to send to the model.
  • auto_accept_trivial_fixes: When true, trivial single-token or formatting fixes can be applied automatically (use with caution).

​
Integration with other AI features

  • Edit Prediction: Use inline suggestions for immediate edits while Edit Prediction offers next-step predictions as you type.
  • Agent Panel: When a task requires multi-file changes or high-risk operations, the Inline Assistant will recommend using the Agent Panel instead.
  • AI Rules: Inline Assistant respects AI Rules — any suggestion that would violate a rule is blocked or redacted.
  • MCP & Tools: For verification (e.g., run tests), Inline Assistant can invoke tools via the Model Context Protocol if allowed by project settings.

​
Privacy & security

  • Default behavior: minimal outgoing context. Only the selected lines and a short surrounding window are sent to the model.
  • Use local-only mode for air-gapped or sensitive projects (see AI Configuration).
  • For cloud providers, the user must configure and permit provider usage. Inline Assistant will not send secrets or files matching AI Rules patterns.
  • Audit logs: enterprise installs can log inline requests and responses for review (subject to redaction policies).

​
Best practices

  • Scope requests narrowly: select the function or block you want help with instead of the whole file.
  • Review suggestions: always review AI edits before applying them, especially for security- or correctness-critical code.
  • Use local-first on proprietary codebases to reduce risk of exfiltration.
  • Combine with linters and tests: treat AI suggestions as first drafts and validate them with existing tooling.

​
Troubleshooting

  • No suggestions appear:
    • Ensure ai.inline_assistant.enabled is true.
    • Check AI provider availability and that credentials are configured (AI Configuration).
    • If using a local runtime, verify it is running and reachable.
  • Suggestions are low quality:
    • Provide more focused context (open related files or select a larger range).
    • Try a different model or increase the confidence threshold.
  • High latency:
    • Enable local_first or choose a lower-latency model.
    • Reduce max_context_lines.
  • Suggestions blocked:
    • AI Rules may be redacting or blocking the required context (check project .oppla/ai-rules.json).

​
Developer notes

  • Extensions can implement custom inline handlers to provide domain-specific suggestions (for example, language-aware autofixes). Prefer structured suggestion outputs (patches, ranges, replacements).
  • Keep outputs small and machine-friendly (JSON patches) to allow reliable application by the editor.
  • Provide dry-run and preview modes for any extension-provided inline actions.
This is a stub intended to resolve missing links and provide immediate guidance. I can add UI screenshots, GIFs for common flows, or full step-by-step examples for specific languages next. Would you like a walkthrough for JavaScript, Python, or another language first?