UtterIDE v0.2.0 Beta for VS Code

AI coding.
You stay in control.

A review-first AI coding agent for VS Code. Choose your model, bring your own key and review every change before it is applied.

Get Beta See how it works

Review-first · Bring Your Own Key · Multi-Database · Local Architecture

queries/get_orders.sql · PostgreSQL 16

DeepSeek V4 Est. cost: 0.03 SEK

1. Prompt

→

2. Inspect

→

3. Model & Cost

→

4. Diff

→

5. QA Review

→

6. Approve

> Optimize this PostgreSQL query Enter ↵

Proposed change: 1 file, 3 additions, 2 deletions · Est. cost: 0.03 SEK

Core Workflow

AI proposes. You review. UtterIDE applies.

A transparent review loop inside your editor. No surprise background writes or hidden telemetry.

01 / UNDERSTAND

AI Proposes

The agent inspects repo files, dependencies, and database schemas to draft surgical code patches.

02 / VERIFY

AI Reviews

A second reviewer model audits the patch for regressions, syntax errors, and SQL safety.

03 / CONTROL

You Approve

Inspect the full VS Code diff, file counts, and exact token costs before making a decision.

04 / APPLY

UtterIDE Applies

Changes are written to disk only with your explicit approval, logged to your local ledger.

Review-First

Nothing changes before you approve it.

UtterIDE lets AI analyze and synthesize changes, but the developer stays in control. Every proposed change is presented as a side-by-side diff with line metrics and cost estimates before anything touches your files.

Proposed change · 3 files +28 / -12 lines

services/order_service.py migrations/004_idx.sql config/db.py

-orders = db.query(Order).all()
+orders = db.query(Order).filter_by(status='active').limit(50).all()
 return serialize_orders(orders)

Estimated run cost: 0.04 SEK

Database Synapse

Your code agent understands the database too.

Connect PostgreSQL, PostGIS, and MySQL. UtterIDE inspects schemas, indexes, and execution plans to solve real database problems safely.

PostgreSQL 16 Connected

PII Masking: Enabled · Guardrails: Strict

Schemas

▾ public

orders (4.2M)

customers (820k)

▸ analytics

▾ gis (PostGIS)

delivery_zones

Prompt:

"Why is this spatial query slow on delivery_zones?"

> Diagnostics & Fix:

1. Execution plan: Sequential scan on delivery_zones (cost=3420..8110)
2. Missing spatial index detected on geometry column
3. Proposed SQL:

CREATE INDEX CONCURRENTLY idx_delivery_zones_geom
ON gis.delivery_zones USING GIST (geom);

Expected query speedup: 240 ms → 2.1 ms

Model Selection

Use the model that fits the task.

Current model selection during beta. Updated continuously as new models and providers release.

DeepSeek V4

1M Context · Everyday coding

Heavy repository scanning
Fast surgical code patches
~0.03 SEK average run cost

Claude 3.7

Sonnet & Opus models

Architecture decisions
Multi-file refactoring
Nuanced edge-case review

Gemini 2.5

2M Context · Flash & Pro

Broad codebase ingestion
Deep documentation reading
Large context analysis

OpenAI GPT-4o

Reasoning & General tasks

General coding help
Established reasoning flows
Broad ecosystem tasks

BYOK Transparency

Your API keys. Your usage. Your cost.

Stored in VS Code SecretStorage. Connect directly to your AI provider and pay wholesale rates without middleman subscriptions or token markups.

Active Provider

DeepSeek

Tokens This Run

32,480

Run Cost

0.04 SEK

Session Total

0.17 SEK

Verification

One AI writes. Another AI reviews. You decide.

A dedicated reviewer model audits code for edge cases and syntax regressions before presenting the diff for human approval.

Step 1 · Builder

Builder Agent

Drafts the patch, resolves dependencies, and formats the diff.

→

Step 2 · Reviewer

Reviewer Agent

Audits logic, edge cases, and database safety before approval.

→

Step 3 · Human

Human Approval

You inspect the verified diff and decide whether to apply.

Architecture

Built for developer control.

01 / LOCAL

Local Context

Project context, workspace indexes, and run logs remain inside your development workspace.

02 / SECRETSTORAGE

BYOK Security

Provider credentials and database passwords stay in VS Code SecretStorage and never touch our servers.

03 / RUN LEDGER

No Black Box

Inspect what was read, proposed, approved, or rejected, with transparent token accounting.

Pricing

Start with UtterIDE Beta.

Free during the beta period. Connect your own API key and pay wholesale rates directly to your chosen provider.

UtterIDE Beta

Free

Full extension functionality during the public beta phase.

Review-first diff approval engine
Multi-database synapses (PostgreSQL, PostGIS, MySQL)
Dual AI QA review workflow
Multi-model selector (DeepSeek, Claude, Gemini, GPT)
Local Run Ledger and cost estimator

Get Beta from Marketplace

Your AI Provider (BYOK)

At Cost

Direct wholesale pricing from your chosen provider. 0% markup.

DeepSeek V4: ~0.03 SEK average per code task
Claude 3.7: Direct Anthropic API billing
Gemini 2.5: Direct Google AI Studio rates
OpenAI: Direct platform token rates
Full visibility into token counts and session costs

Explore Supported Models

FAQ

Frequently Asked Questions

Unlike tools that overwrite workspace files automatically, UtterIDE synthesizes code as a pending diff. You see which files and lines change, along with estimated cost, before approving.

BYOK means you enter your own API key for DeepSeek, Anthropic, Google, or OpenAI. The extension communicates directly with that provider. There is no middleman account and zero token markup.

UtterIDE connects to local or remote PostgreSQL, PostGIS, and MySQL profiles. It inspects schemas, indexes, and execution plans, explains slow queries, and proposes SQL migrations with safety guardrails.

No. UtterIDE has no central backend server. Credentials stay in VS Code SecretStorage on your local machine. Selected code context is sent only directly to the AI provider you configure.

Dual AI Review allows a secondary model to audit code written by the primary agent before it reaches you. It catches edge cases, syntax regressions, and logical flaws.

AI coding. You stay in control.