Proposal: Hermes Agent as a remote orchestration surface for Traycer

[RogerKink6]

Hey Traycer team 👋

Heavy Traycer user here (Claude Code + Codex on a daily basis) and an active user of Hermes Agent, an open-source personal AI agent that runs on a server and talks to you through Telegram, Discord, Slack, and ~20 other messaging platforms.

I'm opening this because I keep hitting the same wall, and I think there's a genuinely exciting resolution for both projects.

The scenario I keep wanting

I'm out walking, an idea hits me. I message my agent on Telegram: "add a dark mode toggle to the landing page, test on mobile, open a PR." My agent, running at home, spins up the work, dispatches it to Claude Code and Codex inside a Traycer epic, waits for both to finish, reviews the diffs, merges the clean one, and pings me back on Telegram with the result.

Today this isn't possible, because Traycer is a desktop app. You have to be at the machine to orchestrate. That's fine for deep work sessions, but it leaves the "ambient" orchestration gap uncovered: the kind of thing you do from your phone between meetings, from a hotel room with only an iPad, or from bed at 11pm when an idea wakes you up.

Hermes is the inverse shape: it's messaging-native. It lives in Telegram/Discord/Slack, runs on a server, and can drive CLIs, spawn subagents, manage long-running work, and schedule cron jobs. If Hermes could talk to Traycer's agent bus, Traycer becomes remotely orchestratable from anywhere, not just from the desktop where the host is running.

Two possible integration levels

I see two levels, and I'd love your input on which makes sense:

Level 1 (lightweight): Hermes as an A2A participant via CLI

Hermes runs on the same machine as the Traycer desktop app and drives it through the existing traycer CLI (creating epics, spawning agents, sending A2A messages via traycer agent send, reading transcripts via traycer agent transcript). No protocol changes needed on your side. Hermes would just need to be a first-class citizen in the A2A bus (today canParticipateInA2A gates non-Claude TUI agents out by default).

Question: is there a documented, blessed path for an external orchestrator to drive Traycer like this? Is there a JSON-RPC/SDK surface we should use instead of shelling out to the CLI?

Level 2 (first-class): Hermes as a harness in the picker

Hermes appears in the harness picker alongside Claude/Codex/OpenCode/Cursor/Grok/Droid/Copilot/Amp. I took a crack at this on a fork (RogerKink6/traycer:feat/hermes-harness). I got the protocol Zod schemas + GUI wiring updated and typechecks/tests passing locally. But I want to be upfront that this didn't actually result in a working integration end-to-end. The harness shows up in schema validation, but it doesn't appear in the provider picker at runtime, and I couldn't get the host to actually launch hermes --tui in a PTY.

My understanding is that a harness needs companion changes on the host side (the compiled binary), which lives outside this repo. So I held off on opening a PR. The open-source changes alone aren't sufficient, and I don't want to waste your review time on something I can't validate works. If there's appetite from your side, I'm happy to clean up the branch, add the protocol version bridges, and hand it off as a starting point. But it'd need the internal companion work to actually function.

Why this is a game-changer for Traycer too

The pitch isn't just "I want my agent to talk to your agent." It's that Traycer's orchestration model becomes reachable from any surface: phone, watch, SMS, whatever. The moment a messaging-native agent can drive it, Traycer stops being a desktop app and becomes an orchestration backend.

Concretely:

• Hermes users (and the community is growing fast) get a compelling reason to install Traycer as their "local heavy compute" layer. Right now, a lot of us are hacking together poor-man's versions of what Traycer already does well.
• Traycer users get a story for "how do I kick off an epic from outside my office?" This is, today, the single most common reason I don't reach for Traycer when I should.

A few things this unlocks once it works:

• Ambient PR review: my agent watches a Traycer transcript on a long task and DMs me on Telegram when a child agent goes off the rails, with a one-tap "intervene" button.
• Cross-epic routing from my phone: "Claude Code is stuck on the auth refactor, spin up Codex on the tests in parallel."
• Scheduled orchestration: cron-driven epic kickoff (e.g. "every morning at 9am, have Claude triage open issues into an epic").
• Multi-agent fleet from a chat window: one Hermes instance can dispatch to N Traycer epics across N machines via Tailscale, all from a single Telegram conversation.

What I'm asking

Not asking you to do the work. Asking:

1. Is the direction welcome? (i.e. is "external orchestrator drives Traycer" a use case you want to support, or does it cut against the product vision?)
2. If yes, which level should I aim at first: Level 1 (CLI/A2A participant, zero protocol changes) or Level 2 (full harness)?
3. Is there a better surface than GitHub for this conversation? A Discord, a design doc template, a maintainer I should tag?

I've read through the protocol schemas, the harness PR history (#78 Grok, #90 Droid, #95 Copilot, #223 Amp), and I have a working understanding of the host/epic/agent model. Happy to jump on a call, write a proper design doc, or contribute code. Whatever unblocks this fastest.

Thanks for building Traycer. Genuinely the best multi-agent orchestration UX I've used.

Jean Desauw
GitHub: @RogerKink6

[pranshugupta54]

Hey Jean - thanks for the thoughtful proposal.

This direction is very much aligned with where we’re going. We’re currently working on both remote-host support and a mobile app, so launching and managing Traycer work away from the desktop is already a priority for us.

We’re also very open to Hermes and will definitely support it as a provider. We’d like to make onboarding simple, including options for syncing your existing Hermes memories.

The end state is close to what you described: you’ll be able to launch work through Telegram, orchestrate agents through A2A, and use Claude Code through your existing subscription.

We also have a Discord where we can continue the conversation: traycer.ai/discord. Thanks for sharing the idea, we’re excited about what this can unlock.

[PROZACH034]

We would love to 2nd this, I was literally looking at this yesterday trying to find a way for my Hermes agents to drive traycer. We would love this too!