v0.122.2

parruda/swarmv0.122.2Oct 6, 2026by mmabrouk

Upgrading?

AI Summary

Enables running Pi agents in-process while keeping Claude Code on Daytona, improves direct connection handling, and adds real liveness checks for cron and workers.

Key Highlights

  • Run Pi agents in-process
  • Send custom connection model IDs and keys through gateway
  • Send Anthropic bearer token for in-process sessions
  • Real liveness for cron and workers
  • Run agents with two replicas

New Features

  • In-process Pi agent execution
  • Direct connection handling
  • Anthropic token support
  • High availability (two replicas)
  • Enhanced runner stream handling

Full Release Notes

## What's Changed
* fix(api): send a custom connection's own model id and key through the gateway by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7367
* chore(deps): bump werkzeug from 3.1.8 to 3.1.9 in /services/runner/images/sandbox by @dependabot[bot] in https://github.com/Agenta-AI/agenta/pull/7370
* chore(deps): bump multidict from 6.7.1 to 6.9.1 in /sdks/python by @dependabot[bot] in https://github.com/Agenta-AI/agenta/pull/7371
* fix(runner): send the Anthropic bearer token on in-process Pi sessions by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7372
* [docs] Update API reference docs by @github-actions[bot] in https://github.com/Agenta-AI/agenta/pull/7368
* docs: document the session attachment upload endpoint with an end-to-end example by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7377
* fix(runner): send direct Gemini connections to Pi's google provider by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7378
* [7334] feat(chart,api): real liveness for cron and the two workers by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7356
* fix(ci): run every rendered-chart test, instead of a list someone must remember by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7383
* docs(reference): document mount file upload for working-directory placement by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7382
* fix(runner): a command that runs too long is stopped and reported to the agent, the turn continues by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7381
* [fix] Let recycled and stopped API workers finish long LLM gateway streams by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7387
* feat: run Pi agents in-process; Claude Code and Codex stay on Daytona by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7386
* fix(gateway): keep Bedrock agents on the direct path, clarify Bedrock gateway errors by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7389
* feat(frontend): buy credit top-ups from the app by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7384
* docs(chart): note the GKE Autopilot 600 s cap on the api grace period by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7391
* fix(api): register the starter-credits connection's LLM gateway endpoint by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7393
* fix(sdk): keep the runner stream open past the runner's tool-call limit by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7395
* fix(runner): keep usage, output and cost on long-run chat spans; price the built-in gateway model by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7396
* fix(runner): export a long run's trace in requests under the ingest size limit by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7402
* docs(guides): add Connect Slack guide with tutorial video by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7397
* blog: open post images at full size on click by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7403
* [release] v0.122.2 by @github-actions[bot] in https://github.com/Agenta-AI/agenta/pull/7369


**Full Changelog**: https://github.com/Agenta-AI/agenta/compare/v0.122.1...v0.122.2

Related Repositories