Skip to main content

AxonFlow v7.1.0 Release Notes

AxonFlow v7.1.0 is the release where several important governance primitives stopped being "platform internals" and became part of the day-to-day developer experience across the SDKs, frameworks, and plugins.

The platform changes are important on their own, but the bigger story is how they change existing integrations:

  • framework users can now explain blocked decisions from the same SDK client they already use
  • step-gated workflow users can control cached versus fresh evaluation and resume from workflow checkpoints
  • coding-assistant plugins can surface richer block context and unlock governed self-service actions through MCP tools

If you are upgrading, read this page together with SDK Version Compatibility, Decision Explainability, and Session Overrides.

Platform additions in v7.1.0​

1. Decision explainability​

The new explainability endpoint lets a caller retrieve the full explanation for a prior policy decision using a decision_id. That response includes:

  • matched policies
  • matched rule detail
  • risk level
  • whether an override is allowed
  • whether an active override already exists
  • a rolling 24-hour session hit count for the same decision pattern

This is documented in Decision Explainability. It matters because a block is no longer a dead end. Framework code, plugins, and operator tooling can all move from "request denied" to "here is why, and here is the governed next step."

2. Session overrides​

AxonFlow now supports governed, time-bounded, audit-logged session overrides on the new /api/v1/overrides surface. Overrides are designed for temporary unblocks with a mandatory justification, server-side TTL clamping, and strict refusal for critical-risk policies.

See Session Overrides for the full model and Approvals And Exception Handling Patterns for when teams should use overrides instead of permanent policy edits.

3. Workflow checkpoints and execution boundary semantics​

For Workflow Control Plane users, v7.1.0 adds explicit retry semantics and checkpoint APIs:

  • retry_policy with idempotent versus reevaluate
  • step-gate response metadata showing whether a decision was cached or freshly evaluated
  • workflow checkpoint listing
  • checkpoint resume operations

This is the biggest framework-facing addition for teams running LangGraph or any external orchestrator through Workflow Control Plane. The details live in Workflow Checkpoints.

4. Risk-tiered approval routing​

v7.1.0 also adds risk-tiered approval routing for approval-driven workflows. Severity is derived from the matched policy risk score or approval configuration, then used to route low-, medium-, high-, and critical-risk requests differently.

In enterprise deployments, that can include:

  • low-risk auto-approval after a configurable delay
  • critical-risk escalation once an SLA is exceeded
  • organization-level risk-tier automation through the dedicated router

The relevant deployment knobs are AXONFLOW_RISK_TIER_ENABLED, AXONFLOW_LOW_AUTO_APPROVE_DELAY_MIN, and AXONFLOW_CRITICAL_ESCALATION_SLA_MIN. For the user-facing governance model, see HITL Approval Gates and Approvals And Exception Handling Patterns.

What changed for SDK users​

All 4 SDKs shipped companion releases on 2026-04-18:

  • Python v6.4.0
  • TypeScript v5.4.0
  • Go v5.4.0
  • Java v5.4.0

Across languages, the new SDK surface is consistent:

  • explainDecision / explain_decision / ExplainDecision
  • audit search filters for decisionId, policyName, and overrideId
  • checkpoint listing and resume methods
  • retry-policy support on step-gate requests

That means existing framework integrations inherit new capability without requiring a new adapter architecture. If your application already uses an AxonFlow SDK client in Gateway Mode or Workflow Control Plane, you can now:

  • explain a deny after the fact
  • search audit logs by decision or override lifecycle
  • resume long-running workflows from explicit checkpoints
  • distinguish cached decisions from fresh reevaluation

What changed for framework users​

LangChain, CrewAI, AutoGen, Lyzr, and LlamaIndex.TS​

These integrations did not get brand-new framework-specific adapters in v7.1.0. Instead, they benefit through the SDK clients they already rely on:

  • use explainability after a block or review event
  • correlate audit search with a decision or override ID
  • if you run the framework through WCP or step-gated orchestration, inspect checkpoint state and retry behavior

See the updated integration pages for LangChain, CrewAI, AutoGen, LlamaIndex.TS, and Lyzr.

LangGraph​

LangGraph is where the execution-boundary changes land most directly. Teams using the LangGraph wrapper or step-gated workflows now get a clearer mental model for:

  • when a repeated step is served from cache
  • when a retry forces fresh evaluation
  • how to list resumable workflow checkpoints
  • how to resume from a safe boundary instead of replaying the whole graph

See LangGraph Integration and Workflow Checkpoints.

What changed for plugin users​

OpenClaw v1.3.0​

OpenClaw got the deepest user-facing change in the first release wave:

  • client.explainDecision(decisionId)
  • client.createOverride(...)
  • client.revokeOverride(...)
  • client.listOverrides(...)
  • richer MCP check responses with decision_id, risk_level, policy_matches, and override context

See OpenClaw Integration.

Claude Code v0.5.0, Cursor v0.5.0, Codex v0.4.0​

The three coding-assistant plugins now surface richer block context in the tool-hook path and can use the agent's MCP server for:

  • explain_decision
  • create_override
  • delete_override
  • list_overrides

That turns a terse block into an actionable workflow: understand the decision, decide whether an override is allowed, and take the governed next step from the same working environment.

See Claude Code, Cursor, and Codex.

Upgrade checklist​

  1. Upgrade the platform to v7.1.0 or later.
  2. Upgrade the SDK that your framework integration uses.
  3. Upgrade the plugin if you use OpenClaw, Claude Code, Cursor, or Codex.
  4. Link blocked-user flows to Decision Explainability and Session Overrides instead of showing a dead-end denial.
  5. If you use WCP, add checkpoint-aware recovery and explicit retry semantics to your runbooks. Start with Workflow Checkpoints.

Next release​

The immediate follow-up patch is v7.1.1, which fixed post-release gaps found in install-and-use E2E across the plugin paths. If you are relying on richer block context, MCP-path explainability, or override behavior from plugins, read that page too.

Last updated on