I’ve been building Agent Console, an Obsidian plugin that runs AI agents in your sidebar, and I haven’t posted about it here before – so this is the proper introduction.
It started as a fork of the excellent Agent Client plugin by @RAIT-09 (Apache-2.0). Agent Client already gave Obsidian a real in-vault agent – @-mention a note for context, ask it to write back. Agent Console keeps all of that and builds on it. Here’s what the fork adds:
Run several agents at once, in tabs. Each chat is its own tab in a single sidebar pane, so you can have one researching, one drafting, and one cleaning up a script – all in parallel. Status icons show whether each is ready, busy, or waiting on you, so you stop babysitting one before starting the next.
Pin multiple notes as lasting context. Instead of just the note you happen to be looking at, pin several notes to a chat and they stay as that conversation’s context as you move around your vault. The chat owns its context – it won’t silently swap when you open a different note.
Long chats stay light on your context budget. Pinned notes are passed by reference, not re-pasted into every message, so a long multi-turn chat uses 65–80% fewer context tokens than re-sending the full notes each turn.
It looks like part of Obsidian. It adopts your theme’s colors and styling instead of hardcoded chrome, so a custom theme restyles it like a built-in panel.
Restart without losing your place. Open tabs and their conversations reopen exactly as you left them.
Built to be dependable. You pin what the agent can see, and edits or commands wait for your OK before they run – nothing happens behind your back. Under the hood it’s held to an enterprise-grade bar: 700+ automated tests, lint and build checks on every change, and a token-efficiency benchmark you can run yourself, so it stays solid as it grows.
It works with the agent you already have – Kiro CLI, Claude Code, Codex, Gemini CLI, or any agent built on the open Agent Client Protocol. Your MCP servers come along, because it runs your agent unmodified.
The current release (1.2.0) adds a smoother first run: on a fresh install it detects an agent you already have, makes it the default, and opens the panel for you. It also cleans up the command palette and adds a one-click path to updates.
Questions and feedback very welcome.
Disclaimer
Is this project open source? Yes (Apache-2.0)
Is this project completely free? Yes
Is this project vibe-coded beyond the author’s ability to comprehend how it works? No – it is vibe-coded, but it’s built and maintained hands-on, with a 700+ test suite and lint/build checks on every change.
Save your go-to prompts and fire them in one click – the kickoffs you retype every session become context-aware chips above the composer, or a ! search right in the message box.
Search your whole chat history – find a past conversation by title or by what was actually said inside it, then reopen it in a tab.
Take a conversation to another agent – switch a chat to a different agent mid-conversation and it carries everything so far along as context.
Your sessions are safe – reopen the panel or restart Obsidian and your tabs come back; undo a closed chat with its full transcript; half-typed drafts survive.
Every agent speaks Obsidian from the first message – a built-in briefing so any agent renders wikilinks, math, and Mermaid diagrams out of the box.
Keyboard-first – every control is keyboard-operable, with hotkeys for switching tabs and firing prompts.
app.js:1 Plugin failure: agent-console Error: Attempting to register an existing view type "agent-client-chat-view"
at t.registerView (app.js:1:2616779)
at t.registerView (app.js:1:2739708)
at fc.onload (plugin:agent-console:203:37813)
Received this on both ver. 1.2 and 2.0.
Also, what does it mean, 110 commits behind Agent Client? I usually like to go for forks that are not too many commits behind.
Since it’s a fork it uses the same view name (would be a breaking refactor), so you’ll need to turn Agent Client off before you try Agent Console. It’s 100-odd commits behind because of feature divergence (Agent Client has a different approach to multi-session compared to Agent Console). Note that it’s also 300+ commits ahead due to the same reason. LMK if you still have trouble after trying the above suggestion.
Nice work. I’ve been running something similar but zero plugin: two agents (a planner and an executor) coordinating through a shared markdown handoff board note, with Claude pointed at the vault folder. Curious how Agent Console handles agents seeing each other’s output. Can one agent read another’s results, or is each sidebar chat isolated?
For the isolation question — Agent Console tabs do not share state natively, so the shared-note handoff you are describing is the practical answer. Both agents get pointed at the same note as context, one writes first, the next picks it up.
Where that pattern gets trickier is when one of your agents lives outside Obsidian entirely. I have been using a plugin called Local Sync that runs an MCP server against your vault — so CLI-based or desktop agents can read and write notes directly as part of the coordination chain. The planner writes a handoff board note, the executor reads it in through MCP, adds its output, and the whole trace ends up in your vault history.
Stacks well with Agent Console running the in-vault side. You end up with agents inside and outside Obsidian both writing to the same note, which makes it a lot easier to inspect what happened between runs.
Thanks! The chat session files contain chat history, so agents are able to see each others’ outputs. I think they could be prompted to coordinate state through their session files, but the shared markdown handoff note is something I use myself too, and it works out pretty well+is cleaner.
