Skip to content
Agents tracked: 501 Releases (7d): 777 Agent status: no reported issuesSince last limit reset: Claude 10d · Codex 8d For AI agents: Markdown Updated Nov 21, 2026

DeerFlow changelog: what's new each month

Every stable DeerFlow release summarised by month: the highlights, new features, improvements, fixes and anything you need to act on. 3 months covered; the current month updates daily.

November 2026 · so far

1 release: 2.2.0

November brought significant security and architectural improvements to DeerFlow, with stricter defaults, encrypted credential storage, and cascading data cleanup for better system integrity.

Action needed
  • Scheduler /api/scheduled-tasks endpoints now return different error formats.
  • Local dev services no longer listen on; use loopback binding.
  • Multi-process gateway requires GATEWAY_WORKERS > 1 configuration.

Highlights

  • Stored credentials are now encrypted at rest using environment-based keys.
  • Security defaults are now fail-closed for delegated web operations.
  • Thread deletion now cascades to remove associated runs and history automatically.
  • Local development no longer exposes services on external network interfaces.
  • Tool authorization now extends to middleware tools in plan mode.

New

  • Environment-based credential encryption at rest.
  • Stricter multi-process gateway requirements (GATEWAY_WORKERS > 1).

Improved

  • Local development binds to loopback only, improving security posture.
  • Thread deletion cascades to clean up runs and related data.
  • Model-bound history remains append-only to prevent payload supersession.
  • Scheduler contract returns standardized errors via /api/scheduled-tasks endpoints.
  • Tool authorization now covers middleware tools in plan mode.

Fixed

  • Fail-closed security for delegated web_fetch and web_capture operations.

September 2026

1 release: 2.1.0

September brought foundational improvements to DeerFlow's infrastructure and execution model, with stronger security guarantees and clearer runtime boundaries for skill management.

Action needed
  • Database checkpoint delta snapshot frequency configuration has moved to a new location.
  • /mnt/skills is now reserved and cannot be used for other purposes.
  • Docker now binds loopback by default; review port binding configurations if affected.

Highlights

  • Trace IDs now included unconditionally in all Gateway HTTP responses for better observability.
  • Memory management is now pluggable through configurable manager classes for custom implementations.
  • Agent tool execution is now verifiable with tamper-evident signatures on each call.
  • Skill packages now have clearer runtime boundaries defined by SKILL.md directories.

New

  • Verifiable agent execution with tamper-evident signatures on tool calls.
  • Pluggable memory management via configurable manager classes.
  • Unconditional trace IDs in all Gateway HTTP responses.
  • Docker loopback binding by default for published entry ports.

Improved

  • Sandbox replica enforcement for E2B as a capacity limit with updated defaults.
  • SKILL.md directory handling clarified as explicit runtime package boundaries.
  • /mnt/skills is now reserved for managed enabled-only skill projections.
  • Database checkpoint configuration relocated for better organization.

June 2026

1 release: 2.0.0

June brought major upgrades to DeerFlow's core capabilities, including persistent run hydration, self-updating custom agents, and expanded model support. The release also enhanced channel integrations and observability features.

Highlights

  • Runs now hydrate from RunStore and persist across interruptions, allowing workflows to resume seamlessly.
  • Custom agents can now persist edits to themselves, enabling self-improving agent behaviour.
  • Users can bind their own Slack and other IM channel connections directly without admin setup.
  • Added StepFun and MiMo reasoning models to the available model options.
  • Real-time token usage streaming for subagents displayed in the header with detailed spans.

New

  • Run hydration from RunStore with persistence across interrupted runs
  • Self-updating custom agents that persist their own edits
  • User-owned IM channel connections for Slack and other platforms
  • StepFun and MiMo reasoning models
  • Community search integration
  • Discord mention-only mode and thread routing
  • Real-time subagent token usage streaming to header
  • Configurable loop detection with per-tool frequency overrides

Improved

  • Trace names now cleaner for leadagent and custom-agent traces
  • Session IDs improved in trace output
  • Discord channel capabilities expanded with new routing options

Summaries are written automatically from the official release notes (full changelog ↗); check the original notes before relying on a detail. DeerFlow: pricing, features and alternatives · All changelogs