Ritemark v1.6.3: One Conversation, Many AI Agents; An Agent Library You Can Talk To
Two ideas drive this release. A conversation should be a durable workspace, not a runtime session, so you can start a thread with Claude and send the next message through Codex without losing the context. And the Agent Library should be a place where you create and fork helpers, not a read-only browser into hidden .claude/ folders. v1.6.3 ships both, plus bundled runtimes that finally remove the "do you have the right CLI installed" question entirely.
One Conversation, Many Runtimes
Until now, picking Claude or Codex was a per-conversation decision. If you started a thread with Claude and wanted to try Codex on the same task, you had to start over. The history was tied to the runtime.
In v1.6.3, the conversation belongs to you. Send the next message through a different runtime and both turns stay in the same thread, readable side by side, referenceable. Every assistant response now carries a small provenance line that tells you which runtime answered and which model. There is no ambiguity about who said what.
The Plan and Edit toggle on Codex moved too. It used to be a property of the conversation, set when you started the thread. It is now a per-turn footer toggle, the same kind of decision you make when you choose between asking a clarifying question and asking for a draft. You decide on each turn, not at the beginning of the day.
This is one of those changes that feels obvious in hindsight. A conversation is a record of thinking. The runtime is just the engine that helped at that moment.
Agent Library: Create and Fork, Not Just Browse
v1.6.0 added the Agent Library so you could see what was sitting in your .claude/ folders. It was a read-only window. v1.6.3 makes it a place where you actually work.
The empty state went from terminal instructions to two buttons: New skill and New agent. Click either and the library scaffolds a fresh markdown file with the right frontmatter, opens it in the editor, and you start writing. No mkdir, no template hunting.
A + affordance appears on every section header for the same reason. If you are looking at the Agents row, the + next to it creates an agent. The intent is obvious from where you click.
Right-click any row and you get the context menu we should have shipped in v1.6.0: Open, Duplicate, Reveal in Finder, Move scope, Delete. Duplicate is the one we use most. Find a skill that is almost what you need, duplicate it, edit the copy. That is how you build a library, by forking from things that already work.
Each row also has a color-coded icon chip now (32×32, eight brand colors, thirty-three Phosphor icons). Icons get auto-assigned from name and description keywords, so a skill called "summarize meeting notes" gets a notes icon and a "code reviewer" gets something terminal-flavored. If the auto-assignment is wrong, override it via frontmatter (icon: and color:). Descriptions are now visible in the rows themselves, with the file path moved to the tooltip.
AGENTS.md and .agents/ are first-class too. If you keep your agent definitions in OpenAI/Codex layout instead of Anthropic's .claude/, the library finds them without configuration.
Launch Chat
The Agent Library is now connected to the AI panel directly. Right-click any agent and pick Launch Chat. The AI panel opens with that agent pinned as hidden context, an indigo chip in the composer shows you which agent is active, and you can just start typing.
The way it works underneath is straightforward. The agent's full instructions get sent once on the first turn. Subsequent turns get a short reminder so you do not burn context re-sending the entire prompt every message. Switch agents mid-thread and an explicit handoff note goes into the conversation, so the new agent knows what was happening before it took over.
The @mention popup got rewired to use the same Pin Agent mechanism. One chip, one source of truth, no "wait, did the @mention actually take effect" guessing.
For anyone who has wished they could just click "talk to this skill" without remembering the right @mention incantation, this is the answer.
Starter Pack on First Run
If your ~/.claude/ is empty, Ritemark seeds four example helpers on first run:
-
skill-creator (Apache-2.0, from Anthropic)
-
outline-from-notes
-
frontmatter-cleanup
-
document-reviewer
Just enough to play with the Library on day one. Seeding only happens once and never overwrites existing files, so if you already have a .claude/ folder, nothing changes.
Bundled Runtimes
v1.6.1 introduced bundled CLIs. v1.6.3 makes them default, with the rough edges sanded off.
Codex and Claude now ship inside the app bundle. No terminal, no npm install -g, no Node version mismatch. Binaries live at …/extensions/ritemark/binaries/agents/<platform>-<arch>/ and the right one is picked automatically.
If you somehow end up with an architecture mismatch (running an Intel build on Apple Silicon, for example), Ritemark detects it and offers Check and Repair actions in Settings. Updates flow through the app update channel, not through your global npm.
Two practical effects. First, fresh installs work without pre-flight setup. Second, you stop debugging "why is claude not found" issues that turned out to be PATH problems.
Selected Text Docks to the Composer
Selecting text in the editor used to push a global banner to the top of the AI panel. In v1.6.3, the selection appears as a tab anchored directly to the composer input card, where you are about to type your prompt.
Dismissing the tab removes the chat reference but does not clear the editor selection, which solves a small but irritating bug from before. The reference is now visually attached to the conversation it belongs to, not floating somewhere else.
Download
v1.6.3 is available for all platforms:
| Platform | File |
|---|---|
| macOS Apple Silicon (M1/M2/M3/M4) | Ritemark-arm64.dmg |
| macOS Intel | Ritemark-x64.dmg |
| Windows | Ritemark-Setup.exe |
Windows users: SmartScreen may show a warning on first run. Click More info then Run anyway. See detailed instructions
Both macOS DMGs are signed with a Developer ID certificate and notarized plus stapled by Apple. If you are already running Ritemark, the Update Center will pick up v1.6.3. New here? Download, open a workspace, and start writing.
FAQ
What's new in Ritemark v1.6.3?
Two headline changes. First, conversations are durable across runtimes, so you can start a thread with Claude and continue it through Codex without losing context, with provenance shown on each turn. Second, the Agent Library moved from browse-only to create-and-fork, with New skill and New agent buttons, a per-row context menu (Open, Duplicate, Reveal in Finder, Move scope, Delete), and color-coded icon chips. Plus Launch Chat, bundled Codex and Claude runtimes, a starter pack on first run, AGENTS.md support, and selected text that docks to the composer.
How does switching runtimes mid-conversation work?
You just send your next message through a different runtime. Both turns stay in the same thread, with a small provenance line on each assistant response that tells you which runtime and model produced it. The Plan and Edit toggle on Codex is now a per-turn footer toggle instead of a per-conversation property.
What's new in the Agent Library?
The library went from read-only to create-and-fork. The empty state shows New skill and New agent buttons. Every section header has a + affordance. Right-clicking a row gives Open, Duplicate, Reveal in Finder, Move scope, and Delete. Duplicate is the primary way to fork an existing skill. Each row also has a color-coded icon chip auto-assigned from keywords, with the description visible in the row and the file path in the tooltip.
What is Launch Chat?
Right-click any agent in the library and pick Launch Chat. The AI panel opens with that agent pinned as hidden context, shown as an indigo chip in the composer. The agent's full instructions are sent once on the first turn; subsequent turns get a short reminder. Switching agents in-thread sends an explicit handoff note. The @mention popup now routes through the same mechanism, so there is one chip and one source of truth.
Does this work without .claude/ folders?
Yes. v1.6.3 scans AGENTS.md at workspace root and any .agents/ directory alongside .claude/. OpenAI and Codex layout conventions work without configuration.
What's in the starter pack?
If your ~/.claude/ is empty on first run, Ritemark seeds four examples: skill-creator (Apache-2.0, from Anthropic), outline-from-notes, frontmatter-cleanup, and document-reviewer. Seeding happens once and never overwrites existing files.
Are bundled runtimes the default now?
Yes. Codex and Claude CLIs ship inside the app bundle, so no terminal, no npm install -g, no Node version mismatch. Architecture-mismatch detection plus Check and Repair actions live in Settings, and updates flow through the app update channel.
What changed about selected text?
Selection from the editor used to appear as a global banner at the top of the AI panel. It now docks as a tab anchored directly to the composer input card. Dismissing the tab removes the chat reference without clearing the editor selection.
Is Ritemark free?
Yes. Using AI agents requires either an Anthropic API key or a Claude.ai subscription for Claude, or a paid ChatGPT subscription for Codex.