Agenta

v0.105.4 Feature

This release adds 3 notable features for engineering teams evaluating rollout.

Published 8d LLM Frameworks

View tool

✓ No known CVEs patched

Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Topics

agent-builder agent-observability agent-orchestration agent-workspace agentic-ai agentic-workflow

+12 more

agents ai-agent ai-agents ai-automation ai-workflow-builder background-agents harness human-in-the-loop llm-evaluation llm-observability mcp self-hosted

Summary

AI summary

Adds hostings run.sh overrides, updates runner SDKs, fixes runner resume handling, blocks all-client-tool Claude runs, and introduces OpenAI‑compatible Pi agents.

Changes in this release

Type	Severity	Summary	CVE
Feature
Feature	Medium	run.sh auto-includes local compose overrides and gains --recreate/--rebuild options run.sh auto-includes local compose overrides and gains --recreate/--rebuild options Source: llm_adapter@2026-07-18 Confidence: high	—
Feature	Medium	Add OpenAI-compatible endpoints for Pi via vault custom connections in agents Add OpenAI-compatible endpoints for Pi via vault custom connections in agents Source: llm_adapter@2026-07-18 Confidence: high	—
Feature	Medium	Add OpenAI-compatible endpoint label and harness-gated picker in frontend Add OpenAI-compatible endpoint label and harness-gated picker in frontend Source: llm_adapter@2026-07-18 Confidence: high	—
Dependency	Low	Bump Daytona SDK to @daytona/sdk 0.198 and OpenTelemetry (OTel) to 2.x Bump Daytona SDK to @daytona/sdk 0.198 and OpenTelemetry (OTel) to 2.x Source: llm_adapter@2026-07-18 Confidence: high	—
Bugfix	Medium	Recognize client-tool settles as a resume in cold replay for runner Recognize client-tool settles as a resume in cold replay for runner Source: llm_adapter@2026-07-18 Confidence: high	—
Bugfix	Medium	Refuse Claude-on-Daytona runs whose tools are all client-kind Refuse Claude-on-Daytona runs whose tools are all client-kind Source: llm_adapter@2026-07-18 Confidence: high	—

Full changelog

What's Changed

[release] v0.105.3 by @github-actions[bot] in https://github.com/Agenta-AI/agenta/pull/5360
feat(hosting): run.sh auto-includes local compose overrides and gains surgical --recreate/--rebuild by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/5353
chore(runner): bump daytona SDK to @daytona/sdk 0.198 and OTel to 2.x by @jp-agenta in https://github.com/Agenta-AI/agenta/pull/5362
[5357] fix(runner): recognize client-tool settles as a resume in cold replay by @ardaerzin in https://github.com/Agenta-AI/agenta/pull/5364
fix(runner): refuse a Claude-on-Daytona run whose tools are all client-kind by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/5366
feat(agents): OpenAI-compatible endpoints for Pi via vault custom connections by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/5345
feat(frontend): OpenAI-compatible endpoint label and harness-gated picker by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/5346
[release] v0.105.4 by @github-actions[bot] in https://github.com/Agenta-AI/agenta/pull/5368

Full Changelog: https://github.com/Agenta-AI/agenta/compare/v0.105.3...v0.105.4

View diff on GitHub

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Share on X Share on Bluesky

Track Agenta

Get notified when new releases ship.

About Agenta

LLMOps platform for prompt management, LLM evaluation, and observability. Build, evaluate, and monitor production-grade LLM applications with collaborative prompt engineering.

Agenta

Summary

Changes in this release

What's Changed

Related context

Related tools

Earlier breaking changes