ContextMeld v0.2.0

ContextMeld v0.2.0 introduces the new product name and native automatic indexing for supported Claude Code and OpenAI Codex session folders.

Changed JSONL session files are processed locally after a short debounce. A manual full index remains available. ContextMeld has no account, telemetry, cloud backend, automatic uploads, or automatic script/MCP execution.

See the README and changelog for verified behavior, platform limits, and installation instructions. Packages are currently unsigned.

AgentHub v0.1.1

AgentHub v0.1.1

AgentHub now provides a complete, reviewable local workflow for moving useful context between Claude Code and OpenAI Codex. It also turns Skills and MCP from read-only inventories into guarded local resource managers.

Highlights

• Populated, isolated demo mode with sessions, projects, memories, skills, MCP and handoff data.
• Claude/Codex session indexing, timelines, literal full-text search and direct result navigation.
• Editable cross-agent context packages with bounded Git context, selected memories and relevant skills; native clipboard copy plus Markdown/JSON export.
• Complete skill-package inspection, copy, duplicate, import/export, trash/restore and safe atomic SKILL.md editing.
• MCP JSON and TOML add/edit/duplicate/enable/disable/remove/copy with secret masking, backups, expected hashes, parse validation and automatic rollback.
• Reviewed GitHub skill marketplace installs with bounded trees, path-traversal protection, conflict detection and no script execution.
• Correct agent detection: only a discovered executable counts as installed; configuration and history are shown as separate signals.
• Refined onboarding, project/activity pages, dark/light themes, compact layouts, dialogs and compatibility matrix.
• Real Tauri screenshots, a 9-second handoff demo and a second skill-copy demo.

Privacy and safety

All indexed history and memories remain local. AgentHub has no account, telemetry, cloud backend or automatic upload. Marketplace traffic occurs only after an explicit browse/inspect action. Transcript commands, downloaded scripts and MCP servers are never executed.

Packaging

The release workflow builds unsigned Windows, macOS and Linux packages. Windows has been smoke-tested locally with isolated synthetic storage. macOS and Linux artifacts are built by GitHub Actions and are not run-tested on this Windows host.

AgentHub v0.1.0

AgentHub 0.1.0 is the first public release of a local-first control center for coding-agent work.

What is included:

• Claude Code and OpenAI Codex session indexing, timelines, projects and universal search
• Local memories with scopes, tags, revisions, trash, import/export and explicit Context export
• Skills discovery with metadata, review, edit, backup, rollback, copy and deliberate marketplace install
• MCP discovery and safe JSON review/edit with secret values hidden
• Dark desktop UI, first-run onboarding and Windows MSI/NSIS installers

Privacy: transcripts stay on the machine, there is no telemetry or account, and marketplace network access happens only after an explicit Browse or Inspect action. See README and SECURITY.md for provider limits and platform notes.