​
Why Effective Context Management Matters

Collaborating effectively with Softgen’s AI, especially on complex projects, hinges on managing what information the AI has access to (its “context window”). Two common pitfalls can hinder progress and increase token usage:

#SymptomRoot Cause & Impact
1The Overstuffed Thread: A single, very long conversation thread accumulates a massive number of tokens (e.g., consistently well over 100k per AI message).The AI struggles to keep track of all details, potentially leading to errors, forgotten instructions, slower responses, or destructive edits. Token costs also escalate.
2The Fragmented Approach: Starting new chat threads for every minor step or question.The AI constantly needs to re-establish project context (e.g., re-opening the same set of files, re-learning project goals), leading to inefficient token use and slower overall progress.

Tokens & AI Performance: Softgen’s AI model offers significant capabilities:

  • Fast completions, even in complex threads.
  • Effective context handling, making it capable with extensive conversations and numerous files.
  • Good intent understanding and fewer off-topic replies.

While the AI is robust, the core principles of context management remain vital. As a general guideline, aim for a token count per AI message that keeps the AI focused and efficient—often in the 20k to 80k token range. Consistently exceeding this by a large margin, especially with many files open, can still lead to the issues described above.

This guide provides a playbook for managing your context effectively.

​
Step-By-Step Playbook for Optimal Context

​
1. Master File Context Management (Your Primary Control)

Softgen’s File Context Management system gives you direct control over which files are included in the AI’s context window. This is the most effective way to optimize token usage and AI performance.

How to Access & Use:

  • Click the gear icon (Agent Settings) in the chat interface.
  • Navigate to the File Context section.

Key Features:

  • View Open Files: See a list of all files currently in the AI’s context.
  • Selective File Closure: Select and close unnecessary files. You can do this one by one or use bulk operations (e.g., “Close Selected”, “Select All” then deselect essentials).
  • Add Files: Use the intelligent file search (“Type to search files…”) to quickly find and add relevant files to the context.
  • Token Optimization: Regularly closing unneeded files significantly reduces token count per message, leading to faster responses and better AI focus.

Strategy:

  • Monitor Token Counts with the AI Context Monitor: Softgen displays the token count for each AI message. On the new Softgen frontend (accessible at https://new.softgen.ai), you’ll also see a visual AI Context Monitor (a stoplight-style indicator) next to the token count. This provides an at-a-glance understanding of your current context health. Use this to gauge context health. While Softgen’s AI is robust, consistently high token counts can still impact performance and clarity. This indicator uses the following thresholds, which also serve as a good mental model if you are on the classic frontend:

    • 🟢 Up to ~80k tokens (Green Zone): Generally a healthy context size for most tasks. The AI should have good focus.
    • 🟡 ~80k - 99k tokens (Yellow Zone): High context usage. The AI is processing a lot of information. Consider proactively closing any non-essential files using the File Context Management UI, or think about whether your current sub-task is nearing completion and if it’s a good time to summarize for a new thread.
    • 🟠 ~100k - 119k tokens (Orange Zone): Very high context usage. AI performance might start to degrade (slower responses, potential for less relevant suggestions). It’s strongly recommended to prune unneeded files via the File Context Management UI. If the task is complex and ongoing, this is a strong signal to use the “Summarize & Restart” pattern.
    • 🔴 120k+ tokens (Red Zone): Extreme context usage. Risk of AI hallucinations, errors, or significantly off-target responses increases sharply. Immediate action is highly advised. Aggressively close all non-critical files. If the AI seems stuck or confused, use the “Summarize & Restart” pattern (you can even ask the AI in Plan Mode to help generate a concise summary if needed) to continue your work in a fresh, focused thread.

    If token counts are frequently in the yellow, orange, or red zones, or if AI performance degrades, your first step should always be to review and prune open files using the File Context Management UI.

  • Working Set: Keep only the files directly relevant to your current task or sub-task open.

  • Core Files: It can be useful to keep essential project files (e.g., main configuration, core modules, App.tsx) open if they provide crucial overarching context, but be mindful of their token cost.

  • AI Assistance: While the UI is primary, you can still ask the AI to Open file X.py or Close all files except Y.md. The AI also attempts to manage context, but your explicit control is more precise.

​
2. Employ the “Summarize & Restart” Pattern for Very Long Threads

Even with the AI’s enhanced capabilities, extremely long and complex threads can benefit from a periodic reset.

  1. Ask the AI for a summary: Summarize our current task, key decisions, progress so far, and any critical unresolved issues in under 1000 tokens.
  2. Copy the summary.
  3. Start a new chat thread.
  4. Paste the summary as your first message in the new thread.
  5. Use the File Context Management UI (gear icon ) to open only the essential files for the next phase of work.

This pattern provides the AI with a concise, relevant starting point in a fresh context, improving focus and efficiency.

​
3. Use an Advanced Troubleshooting Prompt for Error Loops

If the AI gets stuck in a loop of trying and failing to fix an error, use this prompt to reset its approach:

Your recent attempts to fix the error seem to be too narrowly focused on the immediate problem, without fully considering all variables or potential downstream consequences.

Before attempting another fix, please do the following:
1.  Thoroughly review your entire current context, including all open files and our recent conversation.
2.  Generate a comprehensive debug report that includes:
    *   A clear restatement of the core issue and its context.
    *   A list of all solution paths you've attempted so far for this specific issue, noting what happened with each.
    *   An analysis of why those attempts failed.
    *   Any alternative hypotheses or approaches you haven't tried yet.
3.  Based on this full review, state your new proposed solution path and your confidence level (1-100%) in this new path.
4.  For the remainder of this session, please prepend a confidence level (1-100%) to every coding suggestion or significant recommendation you make.

Pro Tip: Save this prompt (and others you find useful) in Agent Settings (gear icon ) > Prompts for quick access!

​
4. Treat the AI as a “Genius but Potentially Overwhelmed Teammate”

  • Provide clear, specific instructions.
  • Actively manage its context using the tools above.
  • Don’t assume it remembers everything from very early in a long conversation, especially if many files are open. The “Summarize & Restart” pattern and focused file context help mitigate this.

​
Quick Prompts Reference & Customization

Remember to save your most-used or customized prompts in Agent Settings (gear icon ) > Prompts!

GoalPrompt ExampleNotes
Check Open Files & TokensWhat files are currently open? How many tokens does the current context involve?Primarily use File Context Management UI for viewing/managing files. Token counts are also shown per AI message.
Summarize & Restart ThreadGive me a concise summary (under 1000 tokens) of our current task, progress, key decisions, and any critical context so I can start a fresh thread.See Playbook Step 2.
Instruct AI to Manage FilesClose all files except 'main.py' and 'config.json'. or Open 'src/utils/helpers.js'.Supplement with the File Context Management UI for precise control.
Advanced Debug Prompt(See Playbook Step 3 for the full prompt)Use when the AI is stuck in an error loop. Adapt as needed.

​
Accessing the Latest Features

Softgen is continuously improving! Many new features, including UI enhancements like the AI Context Monitor, are often rolled out to our new frontend, accessible at new.softgen.ai, before appearing on the classic interface. If you’re looking for the latest capabilities, be sure to check out the new frontend.

​
When Things Go Wrong

Even with the best prompting and context management, AI can sometimes misunderstand or make unintended changes. Here’s how to handle such situations:

  • Reverting Changes: If the AI makes unintended changes, you can often revert to a previous state of your project using Softgen’s GitHub integration. This allows you to step back and re-prompt with clearer instructions.

  • Getting Support for Persistent Issues: If you’re stuck on an error even after trying the troubleshooting prompts and context management techniques detailed in these guides, you can ask the AI to: Output a full summary of the persistent error, with all necessary context for Softgen technical support to evaluate. Then, provide this summary, along with the timestamp of the problematic AI interaction (visible in the chat thread on both new and classic frontends), when you Submit a Support Request.

​
FAQ

Q: With the AI’s improved long context, do I still need to worry about thread length or file count? A: Softgen’s AI significantly enhances its ability to handle longer conversations and more files. However, for optimal performance, cost-effectiveness, and AI clarity, proactive context management is still crucial. Extremely large contexts (many dozens of files or exceptionally long, unfocused discussions) can still tax any model. Using the File Context Management UI and the “Summarize & Restart” pattern for very extensive tasks remain valuable best practices.

Q: Why not start a new chat thread for every single small task? A: While this gives the AI a completely fresh start, it’s often inefficient. You’ll spend tokens and time repeatedly guiding the AI to re-open the same files and re-learn the broader project goals. It’s about balance: group related tasks and development steps within a coherent thread, but use the strategies in this guide to prevent threads from becoming unmanageably large or unfocused.

Q: Does Softgen automatically manage all context or summarize threads for me? A: Softgen’s AI intelligently attempts to manage context based on the conversation. The File Context Management UI provides you with explicit, granular control. The “Summarize & Restart” technique is currently a user-initiated process. Softgen is continuously evolving, so always check the latest release notes for new features that might further automate or assist with context optimization.

Key Takeaway: Actively manage your AI’s focus by using the File Context Management UI (Agent Settings ) to keep only relevant files open. Monitor per-message token counts. For very long or complex endeavors, use the “Summarize & Restart” pattern. These practices, even with a powerful AI model, lead to more efficient, accurate, and cost-effective AI collaboration.