Changelog
Built in the open,week by week.
Nexora is in active development. Here is where we started, what we've shipped, and the road to public launch after final evaluation.
Road to launch
Done You are here Next
Feb 2026
Research & proposal
Problem study, architecture and a 20-week build plan.
Mar 2026
Foundation
VS Code fork rebranded, nexora-core extension, FastAPI engine.
Apr – May 2026
Core engine
Intent, planner, connectors and approvals. First evaluation build.
Aug – Sep 2026
Platform build-out
MCP, analytics, templates, hardening. v1.0 frozen on 12 Sep.
Sep – Oct 2026
Agent & local engine
Streaming agent, terminal, browser, project facts, inline suggestions.
You are hereQ4 2026
Final evaluation
Independent academic evaluation of v1 against its objectives.
Target Q1 2027
Public beta
Signed installer and a wider early-access cohort.
2027
General availability
Paid plans, team features, macOS and Linux builds.
Editor
Suggestions at the cursor
Nexora now offers inline code suggestions as you type, alongside the agent.
- NewInline code suggestions at the cursor (early).
- ImprovedThe agent asks before it calls a connected platform.
- FixedAn unknown command exit status is treated as a failure.
- FixedEach agent reply renders once, without blank space.
Memory · Connectors
Project facts, MCP sign-in and local agents
Nexora remembers standing facts about a project and can hand work to the coding agents you already use.
- NewStanding project facts that every chat and agent turn can recall.
- NewConnect vendor-hosted MCP servers by signing in.
- NewRun Claude Code, Cursor or Cline from chat in an isolated git worktree, accepted file by file.
- ImprovedOne sign-in path for GitHub and Vercel.
- ImprovedProject memory and run results are visible in the panels.
Chat
Model picker, Gemini and the workflow graph
Choose the model per chat, see plans as a graph, and run code steps in your workspace.
- NewModel catalog with a chat model picker, including Gemini.
- NewWorkflow graph view and a template command.
- NewPlan code steps run directly in your workspace.
- ImprovedChat pauses instead of failing when a model is rate-limited.
- ImprovedMarkdown rendering and chat sessions opened in the editor.
Agent
Nexora Browser
The agent can now browse inside the IDE, and you can watch and click along.
- NewA live browser panel that streams the agent's page inside Nexora.
- NewLive connect and disconnect for platforms in Settings.
- ImprovedWeb search prefers Tavily when it is connected.
- FixedThe agent waits for a captcha instead of inventing search results.
Agent
Terminal commands and run modes
The agent can run commands in a live terminal, under rules you choose.
- NewTerminal tool with confirmation and a security denylist.
- NewFour run modes, from ask-every-time to run-everything.
- ImprovedAgent loop raised to 25 turns with a live turn counter and cost ticker.
- NewAuto-apply setting to skip Accept/Reject diffs when you trust the run.
Chat
Streaming replies and reviewable diffs
Replies stream token by token, and every agent edit arrives as a diff you accept or reject.
- NewStreaming chat and agent turns with a Stop control.
- NewAgent file edits preview as Accept/Reject diffs.
- ImprovedSession rail for switching between conversations.
Engine
Local engine and Windows build
Nexora now starts and supervises its own engine on your machine — no server to run.
- NewThe IDE spawns, health-checks and supervises the local engine.
- NewNexora packaged as a Windows IDE product.
- NewChat turns persist across sessions, so Nexora stops forgetting.
- ImprovedPlatforms, Settings, Analytics and Timeline panels redesigned.
Release
v1.0v1.0 — release freeze
The first complete version of Nexora, frozen for evaluation.
- New621 automated tests passing across unit, integration, e2e and security suites.
- New11 live platform connectors and 8 registered MCP servers.
- NewEncrypted token storage; the engine refuses to start without its key.
- ImprovedSoak and restart-recovery testing with zero errors.
Workflows
Templates, timeline and suggestions
Save a plan once and reuse it; travel back through every run.
- NewBuilt-in templates, save-as-template and portable .nexflow files.
- NewTime-travel timeline of past runs.
- NewSuggestions drawn from your own run history.
Cost
Cost analytics
See what each plan will cost before it runs, and what it actually cost after.
- NewPer-step cost estimates before execution.
- NewAnalytics dashboard for spend and runs.
Connectors
MCP connectors and agent-to-agent
Model Context Protocol servers become first-class steps in a plan.
- NewFive MCP-backed connectors.
- NewA2A agent card and outbound delegation.
IDE
Workflow panels, settings and shortcuts
Plans, outputs and settings get dedicated panels inside the IDE.
- NewWorkflow and task output panels.
- NewSettings panel and keyboard shortcuts.
Milestone
First evaluation build
An agent loop that edits your local files, integrated end to end.
- NewAgent loop with session context and local file patches.
- ImprovedIDE and engine integrated for the first evaluation.
Orchestration
Approval engine, GitHub and Vercel
Plans now wait for a human yes, then execute across GitHub and Vercel.
- NewHuman-in-the-loop approvals and a dependency-graph executor.
- NewReal-time progress over WebSocket.
- NewGitHub and Vercel connectors with OAuth sign-in.
- NewRetries, fallbacks, execution history and rollback.
Planner
Decomposition and dependency graphs
Big requests are broken into steps that can run in parallel where possible.
- NewRequirement decomposer and plan generator.
- NewMCP and REST connector clients.
Memory
Workspace memory and intent
Nexora starts to understand both what you ask and the project you ask it in.
- NewWorkspace memory in a single local file.
- NewPlatform search across the registry.
- NewIntent classification for every request.
Milestone
Day one
The first commit: a VS Code fork named Nexora, and an empty engine.
- NewNexora branding across the IDE.
- NewEngine and extension scaffolding.