⚠️ Currently Unstable — Dev Team Is Working on Recovery
The current build may contain bugs or unexpected behavior. Our team is actively investigating and restoring stability. Please use with caution and report any issues on GitHub.
MenteE SWE
UnstableThe autonomous SWE agent that lives in your terminal.
It interprets your task, searches code, calls tools, and delivers verified changes.
Built for real-world engineering
MenteE is not a chatbot. It is an agent that plans, executes, and adapts — with safety guardrails you control.
Autonomous loop
MenteE plans the steps, invokes the tools, reads the output, and decides what to do next — until your task is done or your iteration budget runs out.
Safe by design
Dangerous commands are blocked by default. Destructive operations always require explicit approval. Every action is logged to an audit trail.
6 built-in tools
run_command, read_files, apply_patch, search_code, list_files, finish_task — the complete toolbox for autonomous software engineering.
Guardrails & budgets
Hard limits on iterations, tool calls, tokens, and wall-clock time. Configurable per task, with optional webhook alerts when thresholds are crossed.
Interactive TUI
See every decision as it happens in a terminal UI powered by Ink. Watch tool calls, token usage, and the agent's reasoning live.
Model-agnostic
OpenAI, Anthropic, Google, Mistral, OpenRouter, DeepSeek, Groq. Switch providers with one config change. No lock-in.
How it works
MenteE is not a chatbot. It is an agent that plans, executes, and adapts — with safety guardrails you control.
You describe a task
Type a plain-English instruction in your terminal. MenteE builds a plan using your repo context.
MenteE plans, executes, and adapts
It explores the repo, edits files, runs tests, and self-corrects on failure — choosing the right tools at each step.
You review and approve
Every change is previewed before it's applied. You can approve, reject, or undo any action.
⚠️ Currently Unstable — Dev Team Is Working on Recovery
The current build may contain bugs. Our team is actively restoring stability — please use with caution and report issues on GitHub.
Install in seconds
Single binary. No Docker, no VMs, no complex setup.
Installation
cargo install mentee-swego install github.com/mentee-ai/mentee-swe@latestcurl -fsSL https://get.mentee.ai | shCLI flags
menteeStart interactive TUImentee "task"Run one-shot taskmentee --model <id>Override default modelmentee -y --auto-approveSkip confirmation promptsmentee --no-tuiPlain stdout outputmentee --max-iterations <n>Limit agent iterationsmentee --config <path>Use custom config file$ mentee "fix the failing auth tests"
Agent started. Model: claude-sonnet-4-20250514
Iteration 1/40 · Tools: 0/60 · Tokens: 0/800k
One config file, seven providers
Switch providers without changing your workflow. Add an API key and set your default model.
| Provider | Models | Environment variable | Docs |
|---|---|---|---|
| OpenAI | GPT-4o, GPT-4o-mini, o1, o1-mini | OPENAI_API_KEY | platform.openai.com |
| Anthropic | Claude Opus, Claude Sonnet, Claude Haiku | ANTHROPIC_API_KEY | console.anthropic.com |
| Google AI Studio | Gemini 2.0 Flash, Gemini 1.5 Pro | GOOGLE_API_KEY | aistudio.google.com |
| Mistral | Mistral Large, Codestral | MISTRAL_API_KEY | console.mistral.ai |
| OpenRouter | Any OpenRouter model | OPENROUTER_API_KEY | openrouter.ai |
| DeepSeek | DeepSeek-V3, DeepSeek-R1 | DEEPSEEK_API_KEY | platform.deepseek.com |
| Groq | Llama 3.3, Mixtral | GROQ_API_KEY | console.groq.com |
Safety is not optional
Every command is risk-classified before execution. Dangerous operations require explicit approval. Deletion is never auto-executed.
| Risk level | Behavior | Examples |
|---|---|---|
| safe | Auto-approved (read, lint, git status) | read_files, list_files, git status |
| moderate | Auto-approved (write, install, test) | apply_patch, npm install, vitest |
| dangerous | Requires explicit approval | rm -rf, sudo, DROP TABLE |
| deletion | Never auto-executes | rm -rf /, git reset --hard |
Every action is logged to an audit trail. Configurable guardrails for iterations, tool calls, tokens, and wall-clock time.
Use cases
Fix failing tests
"fix the failing auth tests and make sure they pass — investigate the auth module, identify the issue, apply a patch, and re-run the test suite to verify."
Add a feature
"add a /health endpoint that returns 200 — add the route to the API router, implement the handler, write tests, and verify the build passes."
Debug regressions
"why did the build start failing after the last PR? — analyze the diff, trace the error to the changed code, propose and test a fix."
Refactors
"rename the User service to Account across the API layer — find all references, update imports and usages, run tests to confirm no breakage."
Onboarding
"explain how this repo is structured and where auth lives — read the directory structure, identify auth-related files, and summarize the architecture."
CI / automation
Headless mode for scripted tasks in pipelines — run menteeswe non-interactively for automated fixes, deployments, and maintenance jobs.
Built with modern TypeScript
No heavy frameworks. No unnecessary abstractions. Just the tools that make autonomous agents reliable.
Roadmap
MenteE is actively developed. Here's what we've shipped and what's coming next.
v0.1 — Agent core
Autonomous loop, 6 tools, 7 providers, TUI, safety guardrails
v0.2 — Memory & skills
Project memory, reusable skills, audit trail, cost tracking
v0.3 — Hooks & MCP
Lifecycle hooks, MCP server, worktree isolation, plugin API
v1.0 — Production ready
Stable API, comprehensive docs, enterprise features
Ready to try MenteE SWE?
Install globally, configure a provider, and start tasks in your terminal.