Skip to content

PunamIDE v2.1.2

Latest
Latest

Choose a tag to compare

View all tags
@mandaloriantrader mandaloriantrader released this 28 Jun 11:40
v2.1.2
7a07ba7

PunamIDE v2.1.2

Overview

PunamIDE v2.1.2 delivers a complete rewrite of the inline autocomplete system — modular architecture, dedicated Rust backend commands, proper lifecycle management, and production-hardened provider support.

Website: https://punamide.com
Discord: https://discord.gg/PFp9KWY3eY
X: https://x.com/PunamIDE

What's New

Inline Autocomplete System Upgrade

The inline code completion engine has been completely refactored from a monolithic 230-line block into a modular, production-grade service architecture.

Architecture:

  • New src/services/autocomplete/ module with 9 files — clean separation of concerns
  • Dedicated Rust Tauri commands (call_fim_completion, call_chat_completion_simple) for lower latency
  • Proper Monaco IDisposable lifecycle — no more module-level flags or stale registrations

Features:

  • FIM Support — Fill-in-the-Middle mode for DeepSeek Coder, Codestral, StarCoder2, CodeLlama, and Qwen Coder models
  • Chat Fallback — Tightly constrained chat prompt for all other models (Claude, GPT-4o, Gemini, Groq, etc.)
  • Auto Mode Detection — Automatically selects FIM or Chat based on your active model
  • LRU Cache — 60-entry Map-based cache with 30s TTL for instant repeated completions
  • Suppression Heuristics — No completions in comments, strings, or empty lines without structural context
  • AbortController Cancellation — Previous requests are properly cancelled when you keep typing
  • Post-Processing — Strips markdown fences, meta-text, FIM tokens, and validates output
  • DeepSeek Thinking Models — Properly disables chain-of-thought for v4/R1 models to get direct code output

Settings:

  • New "Inline Autocomplete" section in Settings → Themes & Editor
  • Mode selector: Auto / Force FIM / Force Chat / Disabled
  • Configurable trigger delay (150–800ms)
  • Configurable max tokens (16–512)

Provider Compatibility:

Provider Model Mode
DeepSeek deepseek-v4-flash Chat (thinking disabled)
DeepSeek deepseek-coder-v2 FIM
Mistral codestral-latest FIM
Ollama qwen2.5-coder:7b FIM
Ollama starcoder2:7b FIM
Groq llama-3.3-70b Chat
Gemini gemini-2.5-flash Chat
OpenAI gpt-4o-mini Chat

Cleanup

  • Removed legacy src/utils/inlineCompletion.ts (unused since v2.1.0)
  • Removed 200+ lines of inline completion code from CodeEditor.tsx

Download

Windows Installer (Recommended)

PunamIDE v2.0_2.1.2_x64-setup.exe

Windows MSI (Advanced/Admin)

PunamIDE v2.0_2.1.2_x64_en-US.msi

Installation

  1. Download the installer asset.
  2. Run the installer (upgrades in-place if v2.1.x already installed).
  3. Launch PunamIDE.
  4. Open Settings → Themes & Editor → Inline Autocomplete to configure.
  5. Type in any code file — ghost text appears after 150ms pause.
  6. Press Tab to accept, Escape to dismiss.

Known Issues

  • Windows only (alpha).
  • DeepSeek v4 models require the thinking: disabled parameter — handled automatically.
  • Completion quality varies by model. DeepSeek Coder (FIM) and Codestral give the best results.
  • Very slow networks may exceed the 8s timeout silently (no error shown).

Feedback

Report bugs via GitHub Issues. Include:

  • PunamIDE version (v2.1.2)
  • Windows version
  • Active AI provider and model
  • Steps to reproduce
  • Console errors if visible (F12 → Console tab)

Do not include API keys, tokens, or private source code.

Community

Discord: https://discord.gg/PFp9KWY3eY

Assets 4
Loading