Changelog & Roadmap

What's on npm today, what's being built, and what's planned. Honest status on every item.

Latest on npm: v0.1.1

Where we are

Codehere is an early-stage open source CLI tool. v0.1.1 is on npm today — it works for semantic code search, AI-assisted editing, and task planning on TypeScript/JavaScript projects.

We're building toward a safety pipeline that scans AI instructions for prompt injection and checks generated code for vulnerabilities before it touches your files. That work is in progress — not yet published.

This page marks items honestly: Shipped means it's on npm and tested. In Progress means the code exists but isn't published yet. Planned means we intend to build it — no guarantees on timing.

Phase 1

Foundation (Shipped)

Core safety layer for AI coding. All items here work today.

Semantic code search (index + ask)

Shipped

Local embeddings, SQLite, natural language queries

AI-assisted editing with SAST

Shipped

Security scanning on every generated edit

Task planning

Shipped

Break down goals into steps with affected files

Multi-provider support

Shipped

OpenAI, Cohere, OpenRouter, Ollama

Review gates

Shipped

See diffs before they apply

Local-first storage

Shipped

Per-repo SQLite, no telemetry

Multi-step orchestration

Shipped

Coordinate changes across files

Execution tracing

Shipped

Every operation logged locally

Phase 2

Core UX (In Progress)

Making the working features excellent. Q1 2026.

Conversation history

In Progress

Ask follow-up questions without re-explaining context

Q1 2026

Reference resolution

Next Up

When you ask about a symbol, resolve it to actual code before answering

Q1 2026

Improved error messages

Next Up

Clear, actionable error messages for every failure mode

Q1 2026

Smart command routing

In Progress

Just type what you want — Codehere figures out the right command

Q1 2026

Remove dead code and stubs

In Progress

Clean up placeholder code that shipped in alpha

Q1 2026

Phase 3

Safety Hardening

Making Codehere the safest way to use AI on code. In progress — targeting v0.2.0.

Prompt injection protection

In Progress

Detects and blocks adversarial prompts before they reach the AI — built, not yet published to npm

Safety benchmark (50 patterns)

In Progress

Public, reproducible benchmark across 5 attack categories — built, run it from source

Output validation pipeline

In Progress

Multi-layer validation: prompt injection, SAST, license, bias — built, not yet published

Audit trail

In Progress

Every fix operation logged to .codehere/ — built, not yet published

Hallucination detection

Planned

Detect when AI generates code referencing files/functions that don't exist

Q2 2026

Trust scoring

Planned

Track which operations succeed/fail to build confidence metrics over time

Q2 2026

Phase 4

Scale & Integration

Broader language support, editor integration, larger codebases. Q2-Q4 2026.

VS Code extension

In Progress

Untested alpha code — not yet published or validated with real users

Q2 2026

Multi-file editing

Next Up

Coordinate changes across multiple files in a single operation

Q2 2026

Large repo optimization

Planned

Efficient indexing for 100K+ line codebases

Q2 2026

Python/Go/Rust support

Planned

Expand beyond TypeScript/JavaScript

Q3 2026

REST API

Planned

Programmatic access for CI/CD integration

Q3 2026

Team collaboration

Future

Shared configurations and patterns across team members

2027

Phase 5

Vision: AI Safety Standard

Become the standard safety layer between AI models and codebases.

Model-agnostic safety policies

Future

Define safety rules that apply regardless of which AI model is used

Compliance frameworks

Future

Pre-built policy sets for SOC2, HIPAA, GDPR-relevant development workflows

Enterprise deployment

Future

Self-hosted, air-gapped, SSO, role-based access

Open safety standard

Future

Publish an open standard for AI-assisted code safety that others can adopt

Have Feedback?

We'd love to hear from you! Your feedback helps us prioritize what to build next.

Back to Home