#general
Sep 30October 5, 2026
Hi everyone, we love Archestra. We've been looking for centralizing our company's (https://indahash.com) AI usage in one tool and Archestra has been perfect for it. We're just starting out and already found it useful for user support queries by using knowledge base with Asana connector. Honestly, Archestra looks like a tool we've been looking for more than a year now. It looks so good. We appreciate all the hard work you've done.
However, one of the things that didnt work was an A2A API call with team's token (not personal). We are building an automation for the legal team, the trigger is outside archestra and the goal for it is to make an A2A call to Archestra with team token, because that's most suitable. Such call fails with
MCP tools unavailable: could not connect to MCP Gateway . Agent is shared with the team, MCPs are shared with the team with added service accounts shared with this team. So the setup appears to be correct.I've gone through this with Claude, here are the findings:
This same bug pattern was already fixed elsewhere in the codebase, just never ported over to this path: launch-spec.ts's resolveGatewayToken() branches on actor.kind explicitly — for a "team" actor it calls TeamTokenModel.findTeamToken(teamId) directly, looking the token up by team ID with no userId/membership indirection at all. The plain SendMessage path in a2a-manager.ts has no equivalent branch — it always resolves tokens through a userId, so once a team actor is collapsed to "system" there's no way back to its own token.```
Claude has fixed it apparently (working on tag `platform-v1.4.0-rc.27`), I am attaching a patch here. Please let me know how would you like to proceed with this:
1. You can take this patch and proceed with it in your own pace or implement the fix on your own
2. I can submit a PR
3. Correct me because all the findings are wrong and the problem is in me not being able to use the tool properly :sweat_smile:
Whichever you prefer, I just want to make the life easy for you and us :heart:👀1
3 repliesHi everyone, it's me again 😅 Am I the only one having performance issues with 1.4 RC's UI? We updated accidentally, because Dockerhub's "latest" tag doesnt care about the release being stable or not 🫠 Since then I kept updating it to the most recent RCs (currently rc31), but still the issues persist.
Problems:
1. The spinner in the sidebar keeps rotating for a 2-3 minutes before it stops (screenshot 1). THis is not a problem on its own, but I assume this may be related to some other issues
2. When I click Studio > Agents, nothing happens. Like literally, there is no feedback to the user that it has been clicked. And again after sometimes 30 seconds, sometimes 2 minutes, the Agents page finally load. Or sometimes they dont and nothing ever happens or I see errors (samples attached in screenshot 2 and 3)
3. MCP registry - same thing as with Agents, sometimes it loads, sometimes it doesnt. If it does, it takes a long time.
4. The entire web app feels laggy 😕 Hard to tell exactly, but the UX is just of lower quality than it was on 1.3, especially the menu items.
None of those issues existed on 1.3. I wish we could downgrade to 1.3, but I think there were some DB migrations in between and Archestra won't start after pinning the docker version to 1.3.x. Unless there is some easy way to rollback and maintain all the work we did on 1.4 (creating new agents).
Of course all caches have been cleared, and Archestra is used in a fresh browser icognito session. It sits behind Cloudflare, which also had its Cache purged many times. We only have 9 agents (including users' personal My Assistants) and 10 MCPs (9 self hosted, 1 external). CPU on this machine is at 50% most of the time, RAM at 60%
I am happy to provide whatever additional details might be needed.
Read-only live mirror of Archestra.AI Slack
👋Join the discussion withAI enthusiasts!