AI writes code.
ContextMint decides what ships.

A local-first AI decision-control plane for software teams — not generic chat. Review retrieved evidence before the model answers. Approve patches before anything applies. Code stays on your infrastructure by default.

DECISION-CONTROL PLANE · LOCAL-FIRST · HUMAN OVERSIGHT

In pilot now

v1.0 ships end of August 2026. We are finishing internal pilot readiness; design partner intake opens by start of September 2026. Evidence before inference. Human approval before code changes.

ContextMint Engine

Tray-native operator console — readiness, server, and Ollama status. System, Light, or Dark theme.

Context Lens

Review what will be retrieved and which model runs — before the model answers.

Quality map

Stored governance findings (duplicates, structure signals) with human apply gates.

Image evidence

Paste UI screenshots — local Ollama vision and ARGUS visual audit with human gates.

Default: local Engine + Ollama per developer. What ships in v1.0 → Shared on-prem API server path for regulated pilots — deployment models. Optional BYOK cloud routing is off by default. Roadmap: v1.0 (pilot now) · v2.0 Engineering Intelligence · Track B enterprise & compliance. Full roadmap → · Verification Engine vision → · Getting started →

Your machine, governed

ContextMint Engine is a tray-native operator console — like Ollama or Docker Desktop, not a SaaS dashboard. It follows your OS theme, shows readiness at a glance, and stays out of the way until you need to start, stop, or diagnose. ContextMint is where engineers work. Click any screenshot to enlarge.

Operator console Every tab at a glance — overview, server, models, indexing, and more. Tray-first the rest of the time.
Work surface Readiness → compose → chat → inspector → patch preview. Evidence before inference.
Human gate Summary strip, retrieved evidence, and governance notes before the model runs.

Getting started guide Features (v1.0)

Gate 1 Context Lens — evidence before inference
Gate 2 Patch preview — humans approve applies
Local-first Index and chat on your infrastructure

The Problem Isn't Code Generation. It's Trust.

AI creates changes faster than organizations can review them with evidence. That is a board-level risk — not a tooling gap.

AI Creates
Changes faster than humans can reason about them
Reality is Fragmented
Ownership, intent, and policy live in different tools
Result
Risky merges and ungoverned agent output

The Bottleneck Moved

Code generation is no longer the constraint. Evidence and human oversight are.

Before AI

Humans generated code slower than organizations could review it.

Bottleneck: Code Generation
After AI

AI generates code faster than organizations can verify with evidence.

Bottleneck: Evidence & Governance

What v1.0 delivers

Evidence-based generative chat, store-first governance lookup from the quality map, and grounded recommendations with human approval — on your machine or a guided sovereign server path.

Decision-control plane | Scenario: assisted change in VS Code
Checks before the model answers or applies
✓ Context Lens — retrieved chunks, scores, routing
✓ Quality map — stored findings (duplicates, structure)
✓ Citations — answers grounded in indexed excerpts
⚠ High-severity apply — blocked until human review
✓ Patch preview — unified diff before anything writes
Recommendation
HUMAN APPROVAL REQUIRED
Traceable Evidence
Context Lens chunks
Findings Store finding IDs
Patch preview unified diff
Routing badge local model path
v1.0 decision-control example: Context Lens, quality map, citations, apply gate, and patch preview with human approval and cited evidence.

Roadmap PR verification gates, blast-radius SKU, and CI packaging are the Verification Engine direction — after v1.0, not in the August GA package.

Why CI Tools Don't Solve This Alone

Linting and tests catch syntax. They don't give developers evidence gates in the IDE.

GitHub / GitLab
Code & PRs
No pre-inference evidence review in the IDE
SonarQube / SAST
Syntax & CVEs
No human apply gate for assisted edits
Jira / Confluence
Plans & docs
Disconnected from indexed workspace evidence
ContextMint v1.0
Decision-control in the IDE — Lens, quality map, patch preview, local-first

Govern AI velocity without silent applies

The question is no longer whether AI writes your code.
It is whether humans see evidence before it ships.