LoginSign Up
← Back to all sparks
D

Dify

AI-ASSISTANTS
Velocity2.5

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

dify.ai ↗

Dify pivots from workflow builder to shell-executing agents in a sandbox.

agentsworkflowssandboxskillsself-hostedsecurity
◆Current state
Dify remains an LLM app and workflow platform, but its 2026 releases have steadily shifted weight toward agents. It has added human-in-the-loop workflow nodes, a sandboxed Agent+Skills runtime, and now an experimental Dify Agent that runs in a Linux sandbox and executes shell commands. The patch releases in between (1.14.1, 1.14.2) tightened self-hosting security and workflow reliability around that agent groundwork.
◆Where it's heading
The direction is explicit: Dify is adopting the shell-based, code-executing agent paradigm, with its own preview docs hosted at a bash-is-all-you-need domain. Each release since 1.13.0 has moved from orchestrated workflows toward autonomous agents that run their own tools inside a sandbox, with Skills as the packaging format. The security hardening slotted between feature drops suggests it is readying this for self-hosted production rather than demos.
◆Prediction
Expect 1.16.0 to graduate the experimental Dify Agent toward a stable release, with Skills distribution and sandbox controls as the next areas of investment.

◆Recent moves

  1. 4h ago

    Dify Agent: experimental sandboxed shell agent with Skills

    ⚡ SPARK

    The clearest step yet in Dify's agent pivot: an experimental Dify Agent that runs in a Linux sandbox and executes shell commands, built around Skills. It extends the Agent+Skills runtime introduced in 1.14 into a full agent experience.

    View source ↗
  2. 1mo ago

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

    A patch release consolidating the agent groundwork: tenant-scoped security fixes, workflow and knowledge reliability, and deployment tuning after 1.14.1. Incremental hardening rather than new capability, but the kind of work that precedes a production agent push.

    View source ↗
  3. 1mo ago

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

    Security hardening for self-hosted deployments, notably dropping the public default SECRET_KEY and protecting internal metrics endpoints, alongside workflow stability. Housekeeping that matters most to teams running Dify themselves.

    View source ↗
  4. 4mo ago

    1.14.0-rc1: New Agent x Skills for Production Workflows

    ⚡ SPARK

    The pivot point: Dify's first agent-building runtime, with sandboxed execution, a Skill Editor, and dynamic variable assembly. Everything since, the security patches and the 1.16 Dify Agent, builds on this foundation.

    View source ↗
  5. 4mo ago

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

    ⚡ SPARK

    Human-in-the-Loop added a native Human Input node that pauses a running workflow for review. It closed the gap between fully automated and fully manual flows and set up the collaboration patterns the later agent runtime would need.

    View source ↗
  6. 5mo ago

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

    Summary Index generates AI summaries for document chunks to improve retrieval accuracy on long or scattered content. A real upgrade to the knowledge base, and a reminder that RAG stays part of the platform even as agents take the spotlight.

    View source ↗