Charlieholtz

I'm biased (I'm the creator) but I use Conductor every day. I've recently switched to Opus 4.8 (fast mode always on) as my default model but swap in GPT-5.5 quite a bit for reviewing Opus's work.

My flow is something like: - Create a new workspace for a specific bug/feature - Ramble into the input box. I use a goose neck microphone and Spokenly (with Parakeet as the model ) for local speech-to-text - Hit enter! I don't use plan mode. - Ask for a review from a different model (⌘⇧R) - Create a PR and run a /babysit loop - Run a local version of the app and click around, do a human review. If the LOC are negative we don't pay much attention to the code. If it's positive we do - Merge!

I often have 3-5 workspaces running like this. There's lots of room for improvement but it's been working quite well for me.

Love to hear it! (Conductor founder here).

This is helpful to know - we're working on adding more agents, Copilot and OpenCode harnesses are among the most popular requests.

We also recently built an escape hatch. If you turn on Settings → Experimental → Big Terminal Mode you can create new terminals in the center panel (with ⌘⇧T) and use any agent you'd like (Copilot, OpenCode, etc). It isn't the best experience because you don't get notifications etc (yet), but at least it lets you use the harness you'd like until we build out the first-class UI for it.

Send me feedback anytime, I'm charlie@conductor.build.

Ah shoot sorry to hear that! I’m one of the creators. Can you email me charlie@conductor.build and I can help debug it for you?

co-creator of Conductor here! Conductor is actually a Tauri 2.0 app so it uses the native safari renderer. working on getting remote workspaces working as we speak!

Founding Engineer @ Conductor (San Francisco, ONSITE)

Conductor lets you run a team of coding agents on your Mac. We launched in July through Show HN, and are used by engineers at companies like Linear, Vercel, Notion, Stripe, and YC.

We’re on a mission to build the human interface to AI organizations. We’re a team of four looking for one or two more exceptional engineers or designers. No AI experience needed, just some enthusiasm for the awesome alien power of LLMs.

https://www.ycombinator.com/companies/conductor/jobs/MYjJzBV...

Intel version now live!

Now fixed in the latest version. You can now give Conductor fine-grained GitHub repository access.

Or, skip the integration and use your local GitHub CLI auth.

Fixed! You can now give Conductor fine-grained GitHub repository access.

Or, skip the integration and use your local GitHub CLI auth.

Ah, we’d love for Conductor to feel magical to use — we spend a lot of time thinking about this. Any chance you could say more about what’s lost?

Thank you! Yes I agree :)

Sorry about that! We’re working on making this more intuitive.

Internally our workflow looks like this: - we have a script that sets up a repo — copies env variables, runs pnpm i, inits a db, etc - we have a field in the repo settings called “setup script.” every time you make a new workspace, that script runs

Hopefully will be much improved over the next week or two!

Thank you! Yes, I agree, we're adding plan mode soon. For now I find Opus responds reliably to "plan this out and don't make any changes."

- Entering messages when the agents can result in task lists to save time.

Ah yes, queuing coming soon too! Do you prefer if the queued messages send when Claude is finished or interrupt it?

Yes Mac only (for now)

Let me know how it goes!

We create an isolated git worktree locally on your machine — whereas Codex (I believe) is running a container on the cloud

Claudes' Codes?!

Love to hear it! Your messages are just between you and Claude Code — it all runs on your local Claude Code installation via the SDK (https://docs.anthropic.com/en/docs/claude-code/sdk)

thanks so much!!

Ah thank you! Our designer is Julian Kelly (https://jfk.works). He's amazing, and worked on Messenger at Meta — a background which has been surprisingly helpful since everything we're working on is a spin on chatting.

Yes, we're working on this!

Right now the app uses GitHub's OAuth sign in (https://docs.github.com/en/apps/oauth-apps/building-oauth-ap...) which unfortunately doesn't allow for fine-grained permissions (it will only have access to organization code if you explicitly grant it) . We're switching our sign-in to a GitHub App so we can make the permissions fine-grained.

Hi! Right now the app uses GitHub's OAuth sign in (https://docs.github.com/en/apps/oauth-apps/building-oauth-ap...) which unfortunately doesn't allow for fine-grained permissions.

It will only have access to organization code if you explicitly grant it. We're working on switching our sign-in to a GitHub App so we can make the permissions fine-grained.

wooo that's awesome to hear! keep the feedback coming!

Got it! If you click the repository name in the left sidebar, you should see a field for setup script.

Sorry about that, working on Intel right now!

Ah yeah! Which IDE do you use?

It'd be great to change the default branch used for creating new workspaces. Yeah you can actually change this now! If you click the repo name you can make changes to the "setup script". If you added `git checkout -b "branch name"` it would run that on every new workspace instance.

great post! had never thought of the advantage of binary evals before.

Having binary options forces people to make a decision rather than hiding uncertainty in middle values.

this is great advice that happens to apply to life as well as evals :)

Here are my results:

The Trigger Puller ⬜⬜⬜⬜ Somewhat risk seeking ⬜⬜⬜⬜ Somewhat loss seeking ⬜⬜⬜⬜ Balanced in magnitude sensitivity I earned $192,818.76

played around a bit with emcee and really impressed. giving claude access to APIs and having it just work is really satisfying.

what are next steps?

Just added support for Ollama and LM Studio servers! I'm getting 66 tokens/sec and 0.17s time to first token for Llama 3.1, it's pretty mind-blowing.

https://x.com/charliebholtz/status/1873798821526069258

Only because I haven't tested it on Windows/Linux yet (started working on this last week!). But theoretically should be easy to package for other OS's