LoginSign Up
← Back to all sparks
D

Dify

AI-ASSISTANTS
Velocity1.1

Open-source platform for building, deploying, and operating LLM applications and agents.

dify.ai ↗

Pivoting from RAG app-builder to an agent platform, now stabilizing the surface

llm-app-platformagentshuman-in-the-loopragsecurity-hardeningself-hosted
◆Current state
Dify has spent the last two quarters expanding its capability surface from a workflow/RAG app builder into agent territory: a Human-in-the-Loop node, then a sandboxed Agent runtime with a Skill Editor and collaboration beta. The two most recent releases (1.14.1, 1.14.2) shift register entirely to security hardening, workflow reliability, and self-hosted deployment cleanup, suggesting the new surface is being consolidated rather than extended.
◆Where it's heading
The arc is clear: native human oversight (1.13.0) and agentic execution (1.14.0-rc1) were the directional bets, and the patch releases since are paying down the operational and security debt those features created — tenant isolation fixes, CVE upgrades, Celery/PubSub deployment guidance, and a continued migration to the @langgenius/dify-ui design system. An 'init agent server' commit in 1.14.2 signals the agent runtime is still being built out under the hood.
◆Prediction
Expect a stable 1.14.0 GA that promotes the Agent + Skills experience out of preview, followed by continued agent-server buildout. Near-term patch releases will keep emphasizing security and self-hosted deployment ergonomics.

◆Recent moves

  1. 9d ago

    v1.14.2 - Security fixes, agent groundwork, workflow reliability, and deployment updates

    A patch release continuing the post-1.14.0 consolidation: tenant-scoped endpoint hardening, workflow execution reliability fixes, RAG/knowledge stability, and a quiet 'init agent server' commit that keeps the agent runtime moving. Incremental but broad.

    View source ↗
  2. 16d ago

    v1.14.1 - Security hardening, workflow stability, and cleaner self-hosted deployments

    Security-led patch: a LiteLLM CVE fix, an IDOR fix on the avatar endpoint, hardened internal metrics endpoints, and self-hosted SECRET_KEY handling that drops the public default. Reinforces the deployment-cleanup phase following the agent push.

    View source ↗
  3. 3mo ago

    1.14.0-rc1: New Agent x Skills for Production Workflows

    ⚡ SPARK

    The directional centerpiece: a new sandboxed Agent runtime with a Skill Editor, inline tool calls, dynamic variable assembly, and a collaboration beta. This is Dify repositioning from app builder to agent platform.

    View source ↗
  4. 3mo ago

    1.13.0 - Human-in-the-Loop and Workflow Execution Upgrades

    ⚡ SPARK

    The Human Input node makes human oversight a native part of the workflow graph — pause/resume, review-and-edit, and action-based routing — backed by an execution-engine refactor onto Celery workers. The groundwork that made the later agent runtime credible.

    View source ↗
  5. 3mo ago

    v1.12.0 - Introducing Summary Index: Smarter Retrieval with AI Summarization

    Summary Index adds an AI-generated-summary retrieval layer on top of raw chunk embeddings, with multimodal support and manual editing, aimed at retrieval accuracy on long or scattered documents. A meaningful RAG upgrade within the existing knowledge-base surface.

    View source ↗
  6. 3mo ago

    v1.12.1 - Fix base URL in client and other bugs

    A small bugfix release: client base-URL fix, IRIS data persistence in Docker, and a draft-variable cleanup loop fix. Routine maintenance between feature releases.

    View source ↗