Code generation
Mira is a powerful tool for developers. All models in the family are trained on large volumes of high-quality code and support generation, explanation, debugging, refactoring, and testing across many programming languages.
Supported programming languages
LanguageQualityNotes
PythonExcellentHighest quality, including data science and ML
JavaScript / TypeScriptExcellentReact, Next.js, Node.js, Deno
RustVery goodIncluding async, lifetimes, macros
GoVery goodIdiomatic code, goroutines, channels
Java / KotlinVery goodSpring Boot, Android, enterprise apps
C / C++GoodSystems programming, STL
C# / .NETGoodASP.NET, Unity, WPF
PHPGoodLaravel, WordPress
RubyGoodRails, scripting
SwiftGoodiOS/macOS development
SQLExcellentPostgreSQL, MySQL, SQLite, migrations
Shell / BashVery goodScripting, DevOps
Code generation
Mira can generate code from natural language descriptions. For best results, be specific about requirements and desired style.
Function generation
{
"model": "mira",
"messages": [{
"role": "user",
"content": "Write a Python function that takes a list of strings and returns the most frequently occurring string. If multiple strings have the same frequency, return the first alphabetically. Add type hints and a docstring."
}]
}Result:
Python
from collections import Counter
def most_frequent(strings: list[str]) -> str:
"""Return the most frequently occurring string.
If multiple strings share the highest frequency,
return the first one alphabetically.
Args:
strings: A list of strings to analyze.
Returns:
The most frequent string.
Raises:
ValueError: If the input list is empty.
"""
if not strings:
raise ValueError("Input list cannot be empty")
counts = Counter(strings)
max_count = max(counts.values())
candidates = [s for s, c in counts.items() if c == max_count]
return sorted(candidates)[0]Code explanation
Pass existing code and ask Mira to explain what it does. Especially useful when working with unfamiliar codebases.
Explanation request
{
"model": "mira",
"messages": [{
"role": "user",
"content": "Explain what this code does and where issues might be:\n\nconst debounce = (fn, ms) => {\n let id;\n return (...args) => {\n clearTimeout(id);\n id = setTimeout(() => fn(...args), ms);\n };\n};"
}]
}Debugging
For debugging complex code, we recommend using mira-pro — the model with thinking mode that systematically analyzes code for bugs.
Debugging with thinking
{
"model": "mira-pro",
"messages": [{
"role": "user",
"content": "This Rust code panics on certain inputs. Find the bug:\n\nfn binary_search(arr: &[i32], target: i32) -> Option<usize> {\n let mut low = 0;\n let mut high = arr.len();\n while low < high {\n let mid = (low + high) / 2;\n if arr[mid] == target { return Some(mid); }\n if arr[mid] < target { low = mid + 1; }\n else { high = mid - 1; } // Bug: underflow when mid = 0\n }\n None\n}"
}]
}Tip: when debugging, attach the error message, stack trace, and expected behavior description. The more context, the more accurate the diagnosis.
Refactoring
Mira can suggest improvements for existing code: improving structure, performance, readability, and adherence to best practices.
Refactoring request
{
"model": "mira-pro",
"messages": [
{
"role": "system",
"content": "You are a senior developer conducting a code review. Suggest refactoring with explanation for each change."
},
{
"role": "user",
"content": "Refactor this Express route:\n\napp.post('/users', (req, res) => {\n if (!req.body.email) return res.status(400).send('email required');\n if (!req.body.name) return res.status(400).send('name required');\n db.query('INSERT INTO users (email, name) VALUES ($1, $2)', [req.body.email, req.body.name])\n .then(r => res.json(r.rows[0]))\n .catch(e => { console.log(e); res.status(500).send('error'); });\n});"
}
]
}Test generation
Mira generates comprehensive tests covering main scenarios, edge cases, and error handling.
Test generation
{
"model": "mira-pro",
"messages": [{
"role": "user",
"content": "Write tests for this TypeScript function using Vitest. Cover all edge cases.\n\nexport function parseEmail(email: string): { user: string; domain: string } | null {\n const parts = email.split('@');\n if (parts.length !== 2 || !parts[0] || !parts[1]) return null;\n if (!parts[1].includes('.')) return null;\n return { user: parts[0], domain: parts[1] };\n}"
}]
}Documentation generation
Mira helps generate code documentation: JSDoc/TSDoc comments, READMEs, API documentation, and guides.
Adding documentation
{
"model": "mira",
"messages": [{
"role": "user",
"content": "Add detailed JSDoc comments to every function and type in this file:\n\ninterface Config {\n baseUrl: string;\n timeout: number;\n retries: number;\n}\n\nfunction createClient(config: Config) {\n // ...\n}\n\nfunction fetchWithRetry(url: string, options?: RequestInit) {\n // ...\n}"
}]
}Mira Code CLI
For interactive AI-powered coding, use Mira Code — a CLI tool that integrates with your editor and terminal.
Install and use
# Install npm install -g mira-code # Start interactive session mira-code # Generate code from description mira-code generate "REST API with Express and TypeScript" # Debug a file mira-code debug src/server.ts # Explain a file mira-code explain src/algorithm.rs
Mira Code CLI automatically includes your project context (files, dependencies, git history) for more accurate code generation.
Best practices
- Be specific — specify language, framework, code style, and version — 'Python 3.12, asyncio, type hints'
- Provide context — include existing types, interfaces, and dependencies — the model generates more compatible code
- Review code — always review generated code before using in production
- Iterate — if the result isn't perfect, refine requirements or ask to improve a specific aspect
- For complex tasks — use mira-pro for debugging with thinking mode and mira-max for large codebase work