Skip to main content
Changelog

Changelog

Notable changes to the Agentlify platform, agents, and SDKs.

v1.1.0
Latest

February 2026

New Features

  • Adaptive Clarification Gate (ACG) — Pre-execution check that asks clarifying questions when the user's input is ambiguous, preventing wasted LLM calls on incomplete requests.
  • Context Distillation Layer (CDL) — Between-step summarization that replaces raw output bloat with structured context (facts, open questions, decisions, constraints) for multi-step pipelines.
  • Constraint Verification — Two-phase requirement tracking: extracts constraints from user input, then verifies every step includes a satisfaction section. SHA-256 fingerprinting for drift detection.
  • Dual Persona — Disagreement injection where a skeptic LLM critiques the output and a resolution pass revises it. Configurable scope: final step only or all steps.
  • Per-Step Reasoning — Enable chain-of-thought planning on individual steps instead of globally. Configurable via the Step Editor modal or the API.
  • Optimization Panel — New collapsible UI panel in the Agent Editor (Workflow tab) with toggles for all four optimizations, Dual Persona scope selector, and token cost badges.
  • Clarification Request streaming — New clarification_request SSE event type for ACG questions, supported in both HTTP streaming and callable streaming paths.

Bug Fixes

  • Tool-result resumption — Fixed a bug where buildMessagesForStep dropped role: "tool" messages when resuming after client-side tool execution. The agent now correctly preserves assistant tool_calls and tool-result messages in the conversation history.
  • Callable handler tool passthrough — Fixed the Firebase callable handler not passing tools from the request through to the orchestrator. Affected both streaming and non-streaming callable paths.

Improvements

  • Schema normalization — normalizeSettingsToV2 now merges optimizations with defaults. normalizeStepToV2 migrates legacy global enableReasoning and planningMode: per_step_plan to per-step reasoning objects.
  • Validation — validateStep and validateSettings now enforce types on the new reasoning and optimizations fields.
  • Test suite — 82 new tests across four files covering schema validation, orchestrator optimizations (ACG, CDL, constraints, dual persona, per-step reasoning), context injection, and streaming events.
  • Deprecated — Global enableReasoning toggle in Agent Settings. Reasoning is now configured per-step in the Workflow tab.
Optimization Docs →Custom Tools Docs →
v1.0.0

January 2025

Initial Release

  • Multi-step agent orchestration with planning modes (off, initial_plan, per_step_plan)
  • OpenAI-compatible API — call agents with model: "agent:your-agent-id"
  • Skills system with knowledge retrieval via get_skill tool
  • Custom tools with webhook and client-side execution backends
  • Agentlify JS SDK with automatic tool callback loop
  • SSE streaming with step lifecycle events, tool call visibility, and reasoning token streaming
  • Agent templates, scheduling, and webhook triggers
  • Intelligent model routing via configurable routers
  • Dashboard with agent editor, playground, and analytics

Read: Agent Optimizations

Full docs for ACG, CDL, Constraint Verification, Dual Persona, and per-step reasoning.