#general
Hey I'm Apoorva πβ I recently started getting into open source contributions and Archestra caught my eye, especially the MCP gateway angle. Really cool project.
Tried to open a PR for #3491 (the model selector chip thing) but got blocked by the contributor restriction, so figured I'd say hi here first.
Bit about me:
β’ Software Engineer at Oracle (MySQL / HeatWave release eng β Concourse, Jenkins, Terraform). Before that built a multi-agent platform at a Shopify Plus agency (LangChain, autonomous Dev + QA agents, RAG over Slack/email/meeting data)
β’ Mostly Python + TS day to day, CSAI background.
Took a look at #3491 and have a fix on a branch β tests pass, type-check clean.
Couple of things:
1. What's the move to get unblocked on PRs as a new contributor here?
2. Found a tiny chip flicker on reset I'd want to follow up on separately β bundle or split?
Hi everyone, Iβve seen several issues labeled βReserved for SE interview.β Are you currently hiring? Is this only for Archestra contributors, or can I also participate?
Hi everyone!
Quick one for #2160 β MCP server args now support
$VAR / ${VAR} refs from the env config. No more sh -c wrapping just to interpolate a value. Works for both default and custom YAML deploys. PR up if anyone wants to take a look π
Hey folks! I've been working on Responses API support for the OpenAI provider
(this PR #4143), and I just noticed #4190 is tackling this related portion as well. Rather than let these evolve in parallel, I thought it'd be good to sync up.
Quick summary of what my implementation covers:
- Request/Response/Stream adapters following the existing proxy pattern
- Guardrail integration via toolCalls population for trusted-data evaluation
- unit and e2e test coverage (12 tests + WireMock scenarios)
- extends the current LLM proxy architecture
PR #4190 appears to introduce a broader Model Router architecture that
includes Responses API as one component across multiple providers.
I'm curious in understanding how these pieces could work together or inform each other
@user @user I'm happy to adapt this work based on what makes sense for the architecture, and I'd value your perspective on the best path forward.
Hi all, I'm Justin I have a background in Computer Science with 8 years of experience, some at Amazon Web services, and some as a contractor .
Hey peeps :fox_wave:, i have been looking into proper context compaction. Have some steps to think through and perhaps make multiple PRs as a step by step implementation to get going with context compaction in clear delivery steps.
Hey team!
While working on #4021, I realized we might need a way to move agent configurations between completely different instances (like from dev to prod or sharing with the community).
The basic idea is to export/import portable JSON configs.
I've created a GitHub issue outlining the basic idea - #4201. Let me know what you think!
New release! πππ
1.2.28 (2026-04-29)
Features
- add personal mcp gateway (#4136) (9536b5e)
- add search-and-run tool mode (#4041) (b06cbb7)
- add virtual-key-backed OpenAI-compatible model router (#4190) (6f05365), closes #3851 #720
- allow basic users to access chat agent picker and provider settings (#4188) (24be253)
- default connection page to per-user personal MCP gateway (#4137) (f52d3c1)
- enable
swap_agenttool in Slack/MS teams chatops channels (#4067) (d50bb1a) - *mcp-gateway:* automatic label-based tool assignment (#4070) (8244869)
- *roles:* add user impersonation for role debugging, and export/duplicate (#4160) (9c347a5)
- show 'New messages' label on scroll-to-bottom FAB (#4006) (90643ce)
- split scope and owner filters on agents, LLM proxies, MCP gateways (#4071) (03c02fc)
- support org-level MCP server connections (#3933) (b034cb4)
Bug Fixes
- allow to attach team scoped mcps credentials to org (#4181) (6cdec47)
- allow users to open or download files attached to chat messages (#4192) (74800da), closes #4124
- chat permissions (#4142) (019c7dc)
- complete Okta OIDC and OIN SSO support (#4191) (8ae6f37)
- context too long error not communicated to user for all providers (#3988) (3d9242d)
- guard chat stream onError against AI SDK double invocation (#4047) (c8d5649)
- hide connection page (#4154) (0aa688b)
- MCP server assignment N+1 queries (#4086) (319f7ec)
- prevent table cell overflow on agents (#4182) (ce9fe7b)
- table actions column overflow (#4175) (f921af3)
Miscellaneous Chores
- *backend:* make knip catch dead code (#4178) (4d98830)
- capture raw provider errors as exceptions (#4174) (9b2f8c0)
- capture raw provider errors in Sentry (#4083) (7a8d3c8)
- configurable connection base url meta (#4203) (65cd69d)
- dynamic mcp list (#4193) (c9218ad)
- fix linter after a hotfix (#4157) (210d645)
- gate advanced MCP Gateway / Agent tool UI behind feature flag (#4208) (259e720)
- polish /connection page UX (#4209) (fdad086)
- *release:* bump version (42cf561)
- *release:* bump version (#4143) (c1115dd)
- *release:* bump version (#4155) (8d9609a)
- *release:* bump version (#4186) (c614b4e)
Hello archestra team ,sry to disturb I just wanted to check if there are any issues tagged with a bounty available for contribution
Read-only live mirror of Archestra.AI Slack
πJoin the discussion withAI enthusiasts!Hey peeps :fox_wave:, i have been looking into proper context compaction. Have some steps to think through and perhaps make multiple PRs as a step by step implementation to get going with context compaction in clear delivery steps.
8 replies
excited
@user i could respond to the GitHub issue comments. Maybe π€ we could update the issue with constraints/findings/musts as we go along?
@user hey, sorry i am back now. sorry to hear that it felt like it was ai-generated. let's hope that you got some value from the proposals. sorry if it felt more like noise than signal! π
i want to communicate in clear signals. so appreciate the feedback