This page documents AI assistant systems that are smaller in scope or emerging in the ecosystem. These systems may represent experimental approaches, niche use cases, or newer entrants to the AI coding assistant space. While they share many common patterns with the major systems documented in pages 2, 3, 4, 5, and 6, they often have unique characteristics worth examining.
Currently, this page primarily covers Trae AI, an agentic coding assistant using the AI Flow paradigm. Unlike the extensively documented flagship systems, Trae represents a more streamlined approach to AI-assisted coding.
For comprehensive coverage of major IDE-integrated assistants, see IDE-Integrated Development Assistants. For web-based platforms, see Web-Based Development Platforms.
Sources: Repository structure analysis, Trae/Chat Prompt.txt1-113
Trae AI is an agentic coding assistant that operates within an IDE environment using the "AI Flow paradigm". With an importance score of 3.20, it represents a smaller but distinct system in the repository's collection. Trae emphasizes strict tool schema adherence, precise reply formatting for code edits, and a task evaluation workflow that determines whether direct responses or tool invocations are needed.
Sources: Trae/Chat Prompt.txt1-4
Diagram 7.1.1: Trae System Identity and Operating Model
Trae AI identifies itself as a "powerful agentic AI coding assistant" operating within an "agentic IDE" environment. The system explicitly operates on the "AI Flow paradigm", which enables both independent and collaborative work modes. The relationship with the user is framed as "pair programming" where Trae assists in solving coding tasks.
The task scope encompasses four primary categories: creating new codebases, modifying existing code, debugging, and answering questions. This positioning is similar to other systems like Qoder and Windsurf that also reference the AI Flow paradigm.
Sources: Trae/Chat Prompt.txt1-4
Diagram 7.1.2: Task Evaluation and Response Decision Workflow
The core purpose of Trae's current configuration is to evaluate whether a task requires tool invocation or can be addressed with a direct response. The workflow follows this sequence:
This evaluation step represents a decision layer that determines the mode of operation, distinguishing between tasks that need environmental interaction versus those requiring only informational responses.
Sources: Trae/Chat Prompt.txt6-11
Diagram 7.1.3: Tool System Architecture and Usage Rules
Trae's tool system emphasizes strict adherence to tool schemas and careful tool selection. The configuration file explicitly states "There's no tools you can use yet" at Trae/Chat Prompt.txt18 indicating this may be a template or base configuration.
The toolcall guidelines establish six core principles:
| Guideline | Description | Purpose |
|---|---|---|
| Schema Analysis | Carefully analyze schema definition, strictly follow parameters | Ensure correct tool usage |
| Existence Check | Never call tools not in current tool list (even if in history) | Prevent errors from stale context |
| Tool Exposure | Respond with descriptions, never expose tool information | Security/abstraction |
| IDE Execution | Include call info in response, IDE runs the tool | Delegation pattern |
| Context Analysis | Analyze project, list available tools, select best one | Optimal tool selection |
| Explicit Tools Only | Use only provided tool names, not file/function names | Prevent confusion |
The parameter guidelines add three additional constraints: never fabricate values, use user-provided values exactly (especially quoted ones), and interpret descriptive terms that may indicate required parameter values.
Sources: Trae/Chat Prompt.txt13-38
Diagram 7.1.4: Code Edit Reply Formatting System
Trae has extremely specific requirements for code edit formatting, with particular emphasis on the placeholder syntax. The system mandates:
Critical Placeholder Rule: When showing code edits, Trae MUST use "EXACTLY and ONLY the placeholder // ... existing code ..." to indicate skipped unchanged code. The system explicitly prohibits variations like just "..." or any other format. This exact phrasing appears twice in the prompt at Trae/Chat Prompt.txt44 and Trae/Chat Prompt.txt59 emphasizing its importance.
The edit structure follows this pattern:
cpp:absolute%2Fpath%2Fto%2Ffile
// ... existing code ...
{{ edit_1 }}
// ... existing code ...
{{ edit_2 }}
// ... existing code ...
This approach mirrors similar patterns in systems like VSCode Agent and v0, where unchanged code is represented by markers rather than being fully displayed.
Additional Reply Guidelines:
| Rule | Specification | Location |
|---|---|---|
| Code Block Language | Must match code's grammar, else use plaintext | Line 73 |
| File Paths | Absolute if determinable, else relative from current directory | Lines 60-62 |
| Terminal Commands | Match Windows OS by default, one command per block | Lines 63-72 |
| Comments | Don't modify existing user comments unless asked | Line 74 |
| Project Creation | Create in current directory, not new subdirectory | Line 75 |
| Bug Fixes | Output fixed code, don't just describe fix | Line 76 |
| Images | Use vision capabilities to extract meaningful information | Line 77 |
| Copyright | Avoid infringing content | Line 78 |
| Sensitive Topics | Decline politically sensitive or privacy-related questions | Line 79 |
| Code Runnability | Generated code must run immediately | Line 80 |
| File Rewrites | Only rewrite entire file if explicitly requested | Lines 54, 81 |
| Expertise Scope | Limited to software development topics | Line 82 |
Sources: Trae/Chat Prompt.txt40-83
Diagram 7.1.5: Dual Citation System Architecture
Trae implements two distinct citation systems that must never be mixed:
Web Citation System:
[web:ID] where ID is an index numberCode Reference System (XML Format):
<file_reference filename="$filename"></file_reference><symbol_reference filepath="$filepath" startline="$line" symbolname="$symbolname"></symbol_reference> - startline attribute is required, representing the first line of symbol definition, counting from 1 including blank and comment lines<url_reference linktext="$linktext"></url_reference><folder_reference foldername="$foldername"></folder_reference>The system explicitly warns at Trae/Chat Prompt.txt105-112 that these formats are "entirely separate" and the appropriate format must be used for each context. This prevents confusion between web-sourced information and codebase-internal references.
Sources: Trae/Chat Prompt.txt85-112
| Feature | Trae | Qoder | Windsurf | VSCode Agent |
|---|---|---|---|---|
| Importance Score | 3.20 | 284.89 | 4.48 | 5.20 |
| AI Flow Paradigm | Yes | Yes | Yes | No |
| Tool System | Template (no tools defined) | 8 categories | 7 categories | 5 categories |
| Edit Format | // ... existing code ... | search_replace tool | ReplacementChunk | insert_edit_into_file |
| Parallel Execution | Not specified | Detailed rules | Detailed rules | Not specified |
| Memory System | None | Dual-scope | create_memory | None |
| Citation System | Web + XML references | Markdown links | Not specified | Not specified |
| Code Placeholder | Exact syntax enforced | Context-dependent | Not specified | // ...existing code... |
Trae's configuration suggests it may be:
The emphasis on exact placeholder syntax and dual citation systems indicates attention to output quality and source attribution, which distinguishes it from some other systems that focus more on tool breadth.
Sources: Trae/Chat Prompt.txt1-113 Repository structure analysis
This section is reserved for additional specialized or emerging AI assistant systems as they are added to the repository. The pattern established by Trae suggests that specialized systems may:
For systems that evolve into major platforms, they would be promoted to dedicated sections in pages 2 through 6.
Sources: Repository structure analysis
Refresh this wiki