LoginSign Up
← Back to all sparks
L

LibreChat

AI-ASSISTANTS
Velocity6.3

Open-source AI chat platform supporting multiple model providers and tools.

www.librechat.ai ↗

LibreChat v0.8.8 ships agent interruption and mid-run approval gates — agentic AI with human checkpoints.

agentic-workflowshuman-in-the-loopagent-interruptionmcpopen-source-llm
◆Current state
LibreChat v0.8.8 is addressing the core reliability problem of long-running agents: they can now be interrupted mid-execution, paused for human input via multi-question approval forms, and tracked with live phase cards and reasoning labels. The rc2 release adds the ability to steer or queue follow-up messages during an active agent run, and recovers tool-limited turns rather than failing. The Helm chart releases track each rc for self-hosted deployments.
◆Where it's heading
LibreChat is maturing from an AI chat UI into a production-viable agentic workflow orchestrator. v0.8.7 built the plumbing — MCP integrations, chat projects, shared-link ACLs, OAuth hardening. v0.8.8 adds the controls that make those agents trustworthy in practice: interruptible, observable, and composable with human approval gates. Stateful sessions are marked experimental, which is the next surface to stabilize.
◆Prediction
The next release will likely promote stateful sessions from experimental to stable and expand the multi-question approval form into a configurable checkpoint mechanism, enabling agents to pause, branch, and resume based on human decisions at defined steps.

◆Recent moves

  1. 13d ago

    v0.8.8-rc2

    ⚡ SPARK

    v0.8.8-rc2 builds on rc1's interruption model with optional activity groups, live phase cards showing agent reasoning progress in real time, and reasoning labels. The cumulative effect is a visible, steerable agent runtime — the opposite of the black-box run-to-completion model.

    View source ↗
  2. 13d ago

    chart-2.0.9: Helm chart for v0.8.8-rc2

    Helm chart version bump to track v0.8.8-rc2 for self-hosted Kubernetes deployments. Infrastructure packaging only.

    View source ↗
  3. 1mo ago

    v0.8.8-rc1

    v0.8.8-rc1 introduced the core agent control primitives: mid-run interruption, multi-question approval forms where agents pause for up to four user inputs before resuming, and a redesigned agent UI. These features address the practical gap between autonomous agents and agents that are actually usable in supervised workflows.

    View source ↗
  4. 1mo ago

    chart-2.0.8: Helm chart for v0.8.8-rc1

    Helm chart packaging for v0.8.8-rc1 with updated operator documentation for stateful sessions and background code capabilities.

    View source ↗
  5. 3mo ago

    v0.8.7-rc1

    v0.8.7-rc1 added the MCP infrastructure that v0.8.8 builds on: shared-link ACLs, OpenID role sync, MCP on-behalf-of flows, resumable cleanup, and OAuth hardening. Also added chat projects and skill file authoring tools, expanding the composition surface for agent workflows.

    View source ↗
  6. 3mo ago

    chart-2.0.6: Helm chart bump

    Helm chart version bump to 2.0.6. No feature changes.

    View source ↗