The private AI platform for engineering

Your AI. Your environment. Your control.

Design, build and review software, infrastructure as code and DevSecOps pipelines with AI on Amazon Bedrock, in your own AWS account, commercial or AWS GovCloud (US).

Request a pilot See how it runs

A failing test, then the fix. Topsoil waits for your keystroke. The work sits in its own git worktree.

No GPUs to run. There is no GPU fleet to run. Amazon Bedrock serves the models.
No vendor in the path. Nothing is hosted by us. Inference stays in your account.
No telemetry. Topsoil collects no telemetry. Repositories, plans and documents stay in infrastructure you control.

DesignWeigh the options and settle on an approach before the plan names a file.
SoftwareTake an issue to a pull request in an isolated git worktree, with a test that has to fail first.
InfrastructurePlan and edit Terraform, CloudFormation and pipeline files in git under the same gates.
ReviewA second model checks the change, and a person always makes the merge.

Data sovereignty

Your code stays in your AWS account

Prompts and outputs travel between your machine and the endpoint you configure.

Your machineCode and outputsSource, chats, credentials and tool output stay on the machine that runs Topsoil.

Your RegionCommercial or AWS GovCloud (US)You pick the Region and the model. Requests are signed with your identity.

Your AWS accountAmazon BedrockServes the model in your account. The context library lives in your own S3 bucket.

What data sovereignty means

Where it runs

Terminal, VS Code and the desktop app

Same models. Same approvals. Three places to work.

Terminal

The TUI

The full product in a terminal. Laptop, build server, or locked-down host.

macOS · Windows · Linux

Editor

VS Code extension

A bridge to the Terminal. It shares your open files, then opens a diff.

Ships inside the app · Not via Marketplace

Desktop

Native application

Projects, a library, documents and a built-in Terminal.

macOS · Windows

Topsoil in VS Code on macOS: the agent works in its own git worktree, and the extension opens a side-by-side diff of the three files it changed: the new regression test, the limiter fix and the changelog.

The same work, in VS Code. The diff opens when the turn ends. Your checkout stays put until you merge.

Inside the workspace

You can see what the model was given

Every answer shows its sources.

The working folder, file by file, with nothing implicit.
A token budget you can watch as you go, before it shows up in a bill.
Per-turn cost, shown as the turn happens.

See more of the product

Watch the demo videos

Route Planner · US-GOV Grok 4.7

Working folder9 files

~/Projects/route-planner

src/eta/cache.ts
src/eta/reroute.ts
src/routes/types.ts
src/server.ts
test/eta-cache.test.ts

3 project files · ~462 / 50k tokens1% of budget

this turn2,975 in · 251 out · ~$0.013

The Topsoil desktop app on US-GOV Grok 4.7: an answer explaining stale ETAs in the Route Planner cache, with the working folder and token budget in the Context panel.

The same panel, in the desktop app. The answer cites the line it read.

Software and AI engineering

It makes the change, then checks it

Read the repo. Propose the change. Run the check. Wait for you.

Codebase

Real workspace context

Read and edit files, follow the language server, and search a local index that honours .gitignore.

Execution

Approved shell and Git

Commands and Git run from an allowlist, without a shell, exactly as shown.

Workflow

Multi-model stages

Explore, design, plan, implement, verify, review and document can each use a different model.

Delivery

Backlog to pull request

/task, /goal and /loop plan first, work in an isolated worktree and open a change for review. Topsoil does not merge.

Verification

Proof that the fix works

A new test has to fail without the change before the work counts as done.

Integration

MCP and your systems

MCP servers, per project. Secrets stay on the machine.

Limits

A ceiling on every run

Every run has a ceiling. An unattended run will not start without one.

Scheduling

Exit codes a scheduler can read

A budget stop has its own exit code, distinct from a failed test.

Errors

Failures that say what to do

In the Terminal, a model error says what happened and what to do next.

Close-up of Topsoil reporting the fix: all 5 tests pass, a table lists the three changed files, and a scratch-copy check confirms the new test fails without the change.

Done means proven. The new test fails without the change.

Control

You approve every edit and command

Plan mode stays read-only until you approve. You see the diff and the exact command, then you decide. Where each control lives.

Plan-first by default. Read-only until explicitly released.
Tool risk classes and per-path protections you set per project.
Command allowlists, with commands run without a shell.
Agent features are off until someone turns them on deliberately.
Spend limits. Monthly user and account budgets are checked before each model call.

Review patch · waiting for approval

src/eta/cache.ts (+1 −1)

−  const entry = etaCache.get(route.id);+  const entry = etaCache.get(`${route.id}:${route.revision}`);

Accept once1 Rejectesc Always this path Always this tool3

Close-up of an Allow apply_patch? card in Topsoil: the patch to src/limiter/sliding-window.ts in red and green, with Allow once, Deny and Always allow.

The approval card. The exact lines wait for you.

jordan@route-planner · topsoil

$ topsoil models
Models in us-gov-west-1 (example)

Meta
  us-gov.meta.llama3-3-70b-instruct-v1:0  tools

xAI
  us-gov.xai.grok-4.6             tools · images
  us-gov.xai.grok-4.7             tools · images
* xai.grok-4.7                    tools · images

* default model

Model flexibility

Any model your account enables

Topsoil lists what the account can call. Pick one for a session, a project or a stage.

Your account decides. Enable a model and it appears.
A different model per stage, so cost matches the task.

Defense, regulated work, and anyone with a jurisdiction

For teams that keep engineering context and commercial documents on their side of the network.

Identity

Sign in as you already do

IAM roles and keys, AWS profiles, IAM Identity Center, or a gateway token.

Network

The path you can reach

Public endpoints, custom HTTPS, or PrivateLink and VPC paths.

Off until you turn it on

Capabilities start switched off. Turning one on asks you first.

Settings · Agent features

✓ File edits
apply_patch, write_file, notebook_edit, and saved plans.
✓ Shell commands
run_command, PowerShell, background jobs, and /sandbox.
✓ MCP servers
Start stdio/HTTP/WebSocket servers and their tools and prompts.
✓ Browser automation
Drive a local Chrome/Edge over CDP.

Agent tools can read and modify files, run commands, and start local servers. Only enable for projects you trust.

Artifacts

$ /report Route Planner 2.4 readiness
write document docs/route-planner-2.4-readiness.pdf

Files3

route-planner-2.4-readiness.pdf 3.31 KB
q3-proposal-summary.docx 18 KB
depot-readiness.xlsx 24 KB

Saved into the working folder. PDF, Word, Excel and Markdown.

Beyond engineering

The same privacy, for the rest of the work

Contracts, proposals and campaign plans stay in your library. Topsoil writes the file: PDF, Word, Excel or Markdown.

Proposals and contracts, drafted against your own templates.
Briefs and plans that can use confidential material.
Reports and spreadsheets from PDFs, Office files and scans.
A shared thread for a team, with the files staying in it.

Terminal

The Terminal, when a desktop will not do

Build servers, jump hosts and locked-down workstations get the same models and the same approvals.

Topsoil on Grok 4.7 showing the finished change as syntax-highlighted diffs: the new regression test, the limiter fix in src/limiter/sliding-window.ts and the changelog entry.

Every file, as a diff. See the walkthrough

At a glance

Topsoil features at a glance

Capability	What it means in practice
Private inference	The endpoint you configure. Your IAM policy decides who can call it.
Model selection	Whatever the account enables, per session, project or stage.
Engineering work	Files, approved commands, Git, language servers, browser and documents.
Plan-first	Read-only until you approve. Diffs, allowlists, path checks.
Project context	Projects, attachments and a library in your own storage.
Local search	An optional index that honours `.gitignore`. Source is not uploaded to build it.
Documents	PDF, Word, Excel, Markdown and diagrams, saved in the working folder.
Integrations	MCP, per project. Secrets stay on the machine.
Delivery	`/task`, `/goal`, `/loop`. Isolated worktree, then a pull request. Topsoil does not merge.
Review	Findings anchored to file and line, before the pull request.
Credentials	The AWS profile you set is the one the model call uses.
Sessions	New sessions start fresh. `topsoil sessions scrub` redacts old secrets.
Slack	Off by default. Status only. Never code, diffs or prompts.
Spend	A ceiling per run, and monthly limits checked before each call.
Schedulers	Distinct exit codes for failure and a budget stop.
Collaboration	A shared thread and selected files. Credentials and raw tool output stay put.
Surfaces	Terminal, VS Code, desktop for macOS and Windows, and a web app you host.
Deployment	Your account and region. Nothing is hosted by us.

Questions

Frequently asked questions

Where does Topsoil run?

On Amazon Bedrock, in the account and region you choose. Requests use the endpoint you configure.

Does our source code leave our environment?

No. Prompts, source and documents stay in the account and network you configure. There is no prompt store and no telemetry.

What does data sovereignty mean with Topsoil?

Your code, prompts and outputs stay on the machine and in the account you choose. A cross-Region model can process a request elsewhere in its geography, so pick a model ID that fits. See the data sovereignty guide.

Which AI models can we use?

The models your account enables. The list differs by region. Providers include OpenAI, Anthropic, xAI and Meta, among others.

How do engineers stay in control of changes?

Plan mode stays read-only until you approve. Edits arrive as diffs. Commands need an allowlist. The change comes back as a pull request. Topsoil does not merge it.

Where do engineers use it?

Terminal on macOS, Windows and Linux. A VS Code bridge. A desktop app for macOS and Windows.

Can it run over a private network path?

Yes. Public endpoints, custom HTTPS, or PrivateLink and VPC paths. It needs a reachable endpoint. Sign-in is IAM, an AWS profile, IAM Identity Center, or a gateway token.

Is Topsoil an alternative to GitHub Copilot?

For teams that want a private AI platform for engineering in their own AWS account, yes. See the comparison.

Do we need to run GPUs or a model server?

No. Amazon Bedrock serves the models. There is no GPU fleet to run.

Does Amazon Bedrock use our prompts to train models?

AWS states that Bedrock inputs and outputs are not shared with model providers or used to train base models (AWS). Topsoil keeps no copy.

How do we request a pilot?

Use the form, or email hello@thazen.com. A pilot runs on your code and ends with findings you can act on.

Get started

Try it on your own codebase

A pilot runs on your code and ends with findings you can act on.