GROWBITLABS / PRODUCT 01LOCAL-FIRST AI DEVELOPMENT

STEMCODE

Code faster.Stay in control.

StemCode is a local-first AI coding agent for desktop, terminal, VS Code, Visual Studio, JetBrains, local voice input, and CI code review. It works inside a real repository, uses real tools, and keeps you in control of what runs.

VIEW STEMCODE ON GITHUB ↗EXPLORE STEMCODE →

$npm install -g stemcode

01REAL REPOSITORIES

Search, inspect, edit, validate, and review inside your working tree.

02HUMAN CONTROL

Profiles, permissions, approvals, tracked edits, undo, and redo.

03ONE AGENT, MANY SURFACES

CLI, desktop, editors, ACP integrations, and CI review automation.

███████╗ ████████╗ ███████╗ ███╗   ███╗ ██████╗  ██████╗  ██████╗  ███████╗
██╔════╝ ╚══██╔══╝ ██╔════╝ ████╗ ████║ ██╔════╝ ██╔═══██╗ ██╔══██╗ ██╔════╝
███████╗    ██║    █████╗   ██╔████╔██║ ██║      ██║   ██║ ██║  ██║ █████╗
╚════██║    ██║    ██╔══╝   ██║╚██╔╝██║ ██║      ██║   ██║ ██║  ██║ ██╔══╝
███████║    ██║    ███████╗ ██║ ╚═╝ ██║ ╚██████╗ ╚██████╔╝ ██████╔╝ ███████╗
╚══════╝    ╚═╝    ╚══════╝ ╚═╝     ╚═╝ ╚═════╝  ╚═════╝  ╚═════╝  ╚══════╝

◇ Thinking:

I need to wire the worker host and verify the health endpoint before the next run.

• click to collapse

◆Running apply_patch

• Edited 1 file (+5 −2)

— src/Telemetry.Agent/Program.cs (+5 −2)

8builder.Services.AddHostedService<Worker>();

9 −builder.Services.AddSingleton<IMetricsClient,MetricsClient>();

10var app = builder.Build();

11 −app.MapGet("/health", () => Results.Ok("ok"));

12app.Run();

8 +builder.Services.AddHealthChecks();

9 +builder.Services.AddHttpClient<IMetricsClient,MetricsClient>(client =>

10 +client.BaseAddress = new Uri(builder.Configuration["Metrics:BaseUrl"]!));

11builder.Services.AddHostedService<Worker>();

12var app = builder.Build();

13 +app.MapHealthChecks("/healthz");

14app.Run();

• click collapse Ctrl+click full/preview

◇ Thinking:

Patch is in. Reading the bootstrap file and checking the build output.

• click to collapse

◆Running file read: src/Telemetry.Agent/Program.cs

• Read src/Telemetry.Agent/Program.cs (lines 1–9 of 14)

— preview:

11: using Telemetry.Agent.Services;

22: var builder = WebApplication.CreateBuilder(args);

33: builder.Services.AddHealthChecks();

44:builder.Services.AddHttpClient<IMetricsClient,MetricsClient>(client =>

55:client.BaseAddress = new Uri(builder.Configuration["Metrics:BaseUrl"]!));

66: builder.Services.AddHostedService<Worker>();

77: var app = builder.Build();

88: app.MapHealthChecks("healthz");

99: app.Run();

◆Running dotnet build: src/Telemetry.Agent/Telemetry.Agent.csproj

• Build succeeded in 1.4s

— 0 Warning(s), 0 Error(s)

11: Determining projects to restore...

22: All projects are up-to-date for restore.

33: Telemetry.Agent -> bin/Debug/net8.0/Telemetry.Agent.dll

44: Health endpoint ready at /healthz

◆ Files modified (1)

Updated+5−2src/Telemetry.Agent/Program.cs

BuildDeepseek V4 ProMAX(Deepseek) Thinking Context:1M

500M+LLM tokens processed using StemCode.

SANDBOXED EXECUTIONRun code in an isolated workspace designed to keep project files and execution safely contained.

! AND !! SHELLExecute shell commands directly with ! and quickly rerun or expand commands using !!.

GIT SIDEBARReview changes, track modified files, stage updates, and manage your Git workflow without leaving the editor.

VERSIONABLE MEMORYKeep reusable commands and project memory in reviewable, version-controlled .stemcode/ files.

01 — STEMCODE

Built for practical engineering work, not a detached chat sandbox.

StemCode can understand unfamiliar code, create implementation plans, edit files, run checks, review diffs, and automate pull-request feedback while preserving local control.

Understand

Repository-aware search, file inspection, summaries, symbols, definitions, references, and diagnostics.

LSP + CODE BASE INDEXING + SEARCH

Plan

Turn feature requests and bugs into concrete implementation steps before making changes.

PLAN PROFILE

Edit & validate

Apply tracked edits, run real shell tools, execute checks, and iterate without leaving the working tree.

EDIT, PATCH, SEARCH AND REPLACE AND INSERT LINE

Review

Review local diffs, files, pull requests, and merge requests with a findings-first workflow.

REVIEW PROFILE

F01

LSP-powered code intelligence

Symbols, definitions, references, diagnostics, and rename previews are built into the agent workflow.

F02

Subagents

Delegate focused tasks into independent contexts so the main conversation stays cleaner.

F03

Local voice dictation

Use voice input locally when typing is not the best interaction for the moment.

F04

Shell-first workflows

Run foreground and background terminal commands directly, with path autocomplete for files and directories.

F05

Trajectory Composer

Replay the session as a composed event stream so discovery, commands, edits, and review decisions stay inspectable.

SEE COMPARISONS

Compare StemCode with other coding agents.

Review StemCode against Claude Code, Cursor, GitHub Copilot, and Aider, plus DeepSeek and Harness, across deployment, IDE support, local models, permissions, CI automation, provider choice, and pricing structure.

CLAUDE CODE

StemCode vs Claude Code

A neutral comparison of StemCode and Claude Code across deployment, IDE support, input modes, local models, permissions, CI automation, BYOK, and pricing model.

VIEW TABLE ->CURSOR

StemCode vs Cursor

A neutral comparison of StemCode and Cursor across deployment, IDE support, input modes, local models, permissions, CI automation, BYOK, and pricing model.

VIEW TABLE ->GITHUB COPILOT

StemCode vs GitHub Copilot

A neutral comparison of StemCode and GitHub Copilot across deployment, IDE support, input modes, local models, permissions, CI automation, BYOK, and pricing model.

VIEW TABLE ->AIDER

StemCode vs Aider

A neutral comparison of StemCode and Aider across deployment, editor support, input modes, local models, permissions, CI automation, BYOK, and pricing model.

VIEW TABLE ->DEEPSEEK

StemCode vs DeepSeek

A neutral comparison of StemCode and DeepSeek across deployment, IDE support, input modes, local models, permissions, CI automation, BYOK, and pricing model.

VIEW TABLE ->HARNESS

StemCode vs Harness

A neutral comparison of StemCode and Harness across deployment, IDE support, input modes, local models, permissions, CI automation, BYOK, and pricing model.

VIEW TABLE ->

SEE ALL COMPARISONS ->

02 — ONE AGENT, MANY SURFACES

Use StemCode where engineering already happens.

The same agent experience is designed to move between interactive development, editor workflows, desktop work, and automated code review.

Desktop

Chat, model controls, profile switching, activity output, permission prompts, and undo/redo.

CLI

Keyboard-first sessions, one-shot prompts, piped input, quick reviews, and automation-friendly output.

VS Code

Selected-context prompts, file and diff review, chat, and applying suggestions without leaving the editor.

Visual Studio

A docked StemCode tool window powered by the local CLI through ACP.

JetBrains

An IntelliJ-based IDE tool window using the same local CLI and ACP integration model.

CI reviews

Run automated review workflows in GitHub Actions, GitLab CI, and Bitbucket Pipelines.

03 — SERVICES

Build the missing pieces of your AI stack.

Use GrowBitLabs for one specialized layer or for the whole system, from prototype through private production deployment.

S01↗

Self-hosted LLM systems

Private model deployment, inference APIs, routing, authentication, observability, and operational architecture.

LOCALVPCON-PREMGPU SERVING

S02↗

Custom fine-tuning

Adapt models to approved datasets, domains, structured tasks, and organization-specific workflows.

S03↗

Private RAG

Connect internal files, documentation, databases, and knowledge systems to grounded AI workflows.

S04↗

Edge AI

Optimize and deploy models to laptops, workstations, embedded computers, and offline systems.

S05↗

Custom agent integration

Extend workflows with internal tools, MCP servers, CI systems, policies, and organization-specific automation.

S06↗

Private AI product engineering

Design the full product path from application UX through agents, model routing, data, and infrastructure.

THE GROWBITLABS PRINCIPLE

AI should become
infrastructure,
not dependency.

The model will change. The API will change. Your product should not have to start over every time it does.

04 — BUILT FOR CONTROL

Useful automation without silent surprises.

StemCode makes control visible: profiles separate different kinds of work, permissions define what can run, and sensitive actions can require explicit approval.

PERMISSIONS

Allow, ask, or deny.

Control file edits, shell commands, network access, MCP tools, memory writes, and elevated operations.

PROFILES

Different modes for different work.

Separate implementation, planning, review, exploration, and delegated work.

TRACKED EDITS

Undo and redo agent changes.

Tracked edits include whole-file writes, patches, insertions, and targeted search/replace operations.

WORKSPACE MEMORY

Project knowledge you can review.

Store repeatable commands in .stemcode/commands and long-lived context in .stemcode/memory.

permissions.profileIMPLEMENTATION

File editsASK

Read filesALLOW

Shell commandsASK

Network accessDENY

Memory writesASK

DiagnosticsALLOW

MODEL & PROVIDER FLEXIBILITY

OpenAIChatGPT sign-inAnthropicClaude sign-inGitHub CopilotOpenRouterOpenCode ZenKilo CodeCerebrasGroqDeepSeekGoogle AI StudioOllamaLM StudioOllama CloudStemCode creditsOpenAI-compatible

FAQ

Private AI, without the vague answers.

Are StemCode and GrowBit Convo GrowBitLabs products?

Yes. StemCode and GrowBit Convo are GrowBitLabs products, alongside the broader private AI infrastructure and engineering services shown across the site.

Does StemCode require one specific AI provider?

No. It supports hosted API-key providers, subscription-style sign-in, optional StemCode credits, OpenAI-compatible endpoints, and local model setups including Ollama and LM Studio.

Can StemCode run in CI?

Yes. It supports automated review workflows for GitHub Actions, GitLab CI, and Bitbucket Pipelines.

Can GrowBitLabs help us deploy private models for StemCode?

That is a natural fit for the broader GrowBitLabs service offering: local, VPC, on-prem, or other private model infrastructure can be designed around your workflow.

WORK WITH GROWBITLABS

Bring us the hard AI problem.

Tell us what you want to build, what needs to stay local, which development surfaces matter, and where your models should run.