Agentic Workflow Automation

AEOS now supports fully autonomous workflow execution with configurable human oversight levels:

• Supervised — HITL approval required at every decision point
• Guided — HITL approval required only for high-stakes decisions (configurable thresholds)
• Autonomous — Full automation with NERO quality gates and retrospective audit
• Emergency — Autonomous with circuit breaker and automatic escalation on anomalies

Choose the autonomy level per workflow based on risk tolerance and regulatory requirements.

Advanced NERO Evaluation

NERO now supports:

• Multi-turn coherence — Evaluates consistency across an entire conversation, not just individual responses
• Cross-agent coherence — Ensures outputs from different agents in a workflow don't contradict each other
• Temporal coherence — Detects when agent responses conflict with previously stated facts
• Source attribution — Requires agents to cite specific context sources for factual claims

New Integrations

• WorkOS AuthKit — Streamlined authentication with pre-built UI components
• Stripe — Direct billing integration for SaaS companies reselling AEOS capabilities
• Segment — Event streaming from AEOS to your analytics stack
• PagerDuty — Escalation routing for agent incidents

Looking forward to an exciting 2026 for the AEOS platform.

AEOS Platform v2

The second major version of AEOS is now generally available. v2 brings together all the improvements from the past year into a cohesive platform release:

• UCE v2 with typed schemas and lazy resolution
• MOSAIC v2 with dynamic discovery and cross-workspace orchestration
• AWE v2 with checkpoint/resume on dedicated compute
• NERO GA with custom evaluation dimensions
• Graphiti knowledge graph integration
• Voice agent support
• Visual workflow editor with real-time collaboration

Existing v1 workspaces can upgrade with a guided migration wizard. No breaking changes for agent code — the migration primarily affects workspace configuration.

Prompt Kit

Our prompt engineering framework is now available as a standalone package. Prompt Kit provides a 4-layer architecture for building maintainable, testable system prompts: system layer, context layer, task layer, and guardrail layer. Each layer is independently configurable and version-controlled.

Enterprise Deployment Options

Enterprise customers can now choose between:

• Cloud — Fully managed on ALIAS infrastructure
• Dedicated — Isolated cloud infrastructure with dedicated compute and storage
• Hybrid — Agent execution on your infrastructure, management plane on ours

Multi-Model Support

AEOS agents are no longer tied to a single LLM provider. Configure different models for different agents based on their requirements:

• Use Claude for complex reasoning tasks
• Use GPT-4 for broad general knowledge
• Use Gemini for multimodal processing
• Use smaller, faster models for simple classification tasks

MOSAIC routing considers model cost and latency when assigning tasks. NERO coherence gating works consistently across all supported models.

Agent Testing Framework

The AEOS CLI now includes a testing framework for agent development:

• Define test cases with expected inputs and assertion rules
• Run tests locally against your agent configurations
• Snapshot testing for agent output consistency
• Coverage reports showing which agent skills are tested
• CI/CD integration for automated testing on every commit

ALIAS-ISOF

ISOF (Intelligent Self-Organizing File Meta-System) is now available as an experimental feature. ISOF automatically organizes workspace files, documents, and artifacts using AI-driven classification. Agents can use ISOF to find relevant files without explicit path knowledge.

AWE Runtime on Dedicated Compute

The AWE runtime has been migrated from serverless to dedicated infrastructure. Results:

• Cold starts eliminated entirely
• P95 latency improved by 40%
• Monthly infrastructure cost reduced by 62%
• No execution time limits — workflows run as long as they need

The runtime interface is unchanged. Existing deployments migrated transparently with no code changes required.

Feature Flags with Bucket

AEOS now integrates with Bucket for feature flag management. Roll out new agent capabilities gradually, A/B test different agent configurations, and kill-switch problematic features instantly. Feature flags work at the agent level, workflow level, or individual skill level.

Enhanced Monitoring

• Real-time agent health dashboard with CPU, memory, and network metrics
• Custom alert rules with notification via email, Slack, or webhook
• Agent execution flamegraphs showing time spent in each phase
• NERO coherence score heatmaps across agents and time periods

Graphiti Knowledge Graph

AEOS now integrates with Graphiti for building and querying knowledge graphs that agents can use as long-term memory. Unlike session-scoped UCE context, Graphiti knowledge persists across all sessions and grows as agents process more information.

Agents can query the knowledge graph for facts, relationships, and patterns that inform their responses. The graph updates automatically as agents encounter new information — with NERO validation ensuring only accurate information enters the graph.

Agent Memory System

Beyond Graphiti for factual knowledge, agents now have a structured memory system:

• Working memory — Short-term context within a single execution
• Session memory — Persists across turns within a conversation
• Long-term memory — Selective storage of important facts and patterns via Graphiti
• Organizational memory — Shared knowledge accessible to all agents in a workspace

Enterprise Compliance

New features for regulated industries:

• Data residency controls — specify which regions your data can be stored and processed in
• PII detection and redaction in agent inputs and outputs
• Configurable data retention policies with automatic purging
• SOC 2 Type II audit report available for Enterprise tier customers

CopilotKit Integration

AEOS now integrates with CopilotKit for building copilot experiences on top of your agent workflows. Embed agent-powered copilots in your application's UI with contextual suggestions, inline actions, and conversational interfaces. The copilot layer connects to AEOS agents through MOSAIC, inheriting all orchestration and quality guarantees.

AI-Powered Analytics

The analytics dashboard now includes AI-generated insights:

• Automatic anomaly detection with natural language explanations
• Agent performance recommendations based on usage patterns
• Cost optimization suggestions with projected savings
• Workflow efficiency analysis identifying bottleneck steps

MOSAIC Protocol v2

The orchestration protocol has been upgraded:

• Dynamic agent discovery — Agents register capabilities at startup and MOSAIC routes accordingly. No static routing configuration required.
• Cross-workspace orchestration — Agents in different workspaces can participate in the same workflow (with appropriate auth)
• Priority queuing — High-priority tasks jump the queue when agents are at capacity
• Circuit breaker — Automatic circuit breaking for agents with elevated error rates

Voice Agent Support

AEOS now supports voice-enabled agents through ElevenLabs and LiveKit integration. Build agents that listen, speak, and interact in real time. Voice agents integrate with the same MOSAIC orchestration, UCE context, and NERO quality layers as text-based agents.

Use cases: customer service voice bots, meeting assistants, accessibility-focused interfaces.

Real-Time Workflow Collaboration

Multiple team members can now edit the same MOSAIC workflow simultaneously in the visual editor. Changes sync in real time via Convex subscriptions. Cursor presence shows who's editing which node. Conflict resolution handles concurrent edits to the same node gracefully.

UCE v2

The Unified Context Engine has been substantially upgraded:

• Typed context schemas — Define context types with Zod schemas, get compile-time validation
• Context composition — Compose context from multiple sources with merge strategies
• Lazy resolution — Context fields are resolved on-demand, reducing upfront resolution cost
• Context versioning — Track how context evolves across a workflow's execution

Human-in-the-Loop Approval Workflows

MOSAIC now supports HITL checkpoints as first-class workflow primitives. Insert approval gates at any point in a workflow — the execution pauses, notifies the designated reviewer, and resumes when approved. Reviewers can approve, reject, or modify the agent's output before the workflow continues.

Configurable escalation rules automatically route approvals based on risk level, dollar amount, or customer tier.

Agent Skill Marketplace

Browse and install community-contributed agent skills alongside the curated ALIAS Toolbox. Skills are reviewed for quality and security before listing. Revenue sharing available for skill authors.

Advanced NERO Configuration

NERO now supports custom evaluation dimensions beyond the default coherence, grounding, and alignment. Define domain-specific quality criteria — regulatory compliance checks, brand voice adherence, technical accuracy — and NERO evaluates outputs against your custom rubric.

AEOS CLI v2

The command-line interface has been rebuilt:

• aeos deploy — Deploy agents and workflows from your local project
• aeos logs — Stream real-time logs from any agent or workflow
• aeos test — Run agent test suites locally before deployment
• aeos workspace — Manage workspace configurations and secrets
• Device authorization flow via WorkOS — no passwords in the terminal

Mastra Integration

AEOS now integrates natively with the Mastra agent framework. Define agents using Mastra's TypeScript agent primitives, and AEOS handles orchestration (MOSAIC), context (UCE), quality (NERO), and runtime (AWE). This is our recommended path for TypeScript-first agent development.

Workspace Templates

New workspaces can start from pre-configured templates:

• Customer Operations — Support agent, escalation workflow, NERO quality checks
• Document Processing — Ingestion, OCR, extraction, validation pipeline
• Data Analysis — Query agent, visualization agent, report generation workflow

Templates include agents, workflows, context schemas, and NERO configurations. Customize after deployment.

AWE Runtime v2

The AWE runtime has been rebuilt with first-class support for long-running workflows:

• Checkpoint/resume — Workflows automatically checkpoint state at configurable intervals. If interrupted, they resume from the last checkpoint rather than starting over
• Parallel execution — Fan-out to multiple agents with configurable concurrency limits and aggregation strategies
• Resource isolation — Each workflow execution gets an isolated resource budget, preventing runaway workflows from affecting other workloads

UCE Graph Context

UCE now supports graph traversal queries powered by Neo4j. Instead of flat key-value context, agents can request relationship-aware context: "Get customer X and all their related projects from the last 6 months." The graph context resolves relationship paths and returns structured, typed results.

Enterprise SSO

WorkOS SSO integration now supports:

• SCIM provisioning for automatic user lifecycle management
• Directory sync with Azure AD, Okta, and Google Workspace
• Custom SAML configurations for organizations with non-standard identity providers

ALIAS Toolbox

Toolbox is live — a curated library of pre-built agent skills for document processing, data analysis, communication, and integration. Install skills into your workspace and start using them immediately. Each skill includes a defined interface, dependencies, and NERO coherence expectations.

Semantic Caching

Agent responses can now be cached semantically. When a new query is semantically similar to a previously cached query (above a configurable threshold), the cached response is returned without invoking the agent. Reduces LLM costs by 30-60% for repetitive query patterns.

NERO GA

NERO coherence gating is now generally available across all paid tiers. Since beta, we've improved evaluation speed by an additional 25% and added support for multi-language output evaluation. NERO is enabled by default on new workspaces with sensible default thresholds.

AEOS Vault

Vault is now available for managing secrets across agent systems. Features include encryption at rest (AES-256), scoped access per agent and per operation, automatic credential rotation for supported services, and full audit logging of every secret access.

Reference secrets in agent skill manifests by name — the AWE runtime resolves them at execution time.

Billing Dashboard Improvements

• Real-time cost tracking with per-minute granularity
• Projected monthly cost based on current usage trajectory
• Cost anomaly detection with configurable alert thresholds
• Downloadable invoices in PDF format

MOSAIC Conditional Routing

Workflows now support conditional branches. Route tasks to different agents based on input properties, context values, or the results of previous steps. Conditions are defined using a simple expression language or TypeScript predicates.

Agent Identities (Beta)

Agents can now maintain persistent identity across sessions. Each identity includes state storage, behavioral profiles, reputation scores, and interaction history. MOSAIC uses reputation scores for intelligent task routing — higher-reputation agents get priority for high-stakes tasks.

Visual Workflow Editor

Build MOSAIC orchestration flows by dragging and dropping agent nodes. The visual editor generates standard MOSAIC workflow definitions — everything you build visually can also be defined in TypeScript. Includes live execution preview against sample data.

Langfuse Trace Enrichment

Traces now include richer metadata: UCE context resolution details, MOSAIC routing decisions, NERO coherence scores, and cost attribution. The enriched traces make it possible to debug complex multi-agent workflows end-to-end within Langfuse.

AEOS Playground

The Playground is live — a browser-based environment for testing agent configurations, context schemas, and orchestration flows without deploying to production. Build, test, and iterate in real time. Share configurations via URL.

Webhook Integrations

AEOS now supports inbound and outbound webhooks. Trigger agent workflows from external events (form submissions, CRM updates, Slack messages). Send agent results to external endpoints. All webhook deliveries include retry logic and delivery confirmation.

NERO Beta

NERO coherence gating graduates from alpha to beta. Key improvements since alpha:

• 40% faster evaluation with no accuracy loss
• New "revision" mode that sends below-threshold outputs back to the agent for improvement rather than blocking
• Per-field coherence scoring for structured outputs
• Dashboard integration showing coherence score trends over time

MOSAIC Workflow Versioning

Workflows now support versioning. When you modify a workflow configuration, the previous version is preserved. Roll back to any previous version with one click. Active executions continue on their original version — new executions use the latest.

This enables safe iteration on production workflows without risking disruption to in-flight operations.

UCE Context Snapshots

You can now take snapshots of UCE context state for debugging and testing. Snapshots capture the exact context an agent received at a specific point in time, making it possible to reproduce and debug agent behavior deterministically.

Error Handling Improvements

• Agent errors now include structured error codes with documentation links
• MOSAIC retries failed agent actions with configurable backoff
• AWE runtime provides graceful degradation when downstream services are unavailable
• Dashboard error views now show the full error chain across the agent workflow

NERO Alpha

NERO (Neuro-Emergent Recursive Orchestration) is now available in alpha for all paid tiers. NERO sits between agent outputs and end users, evaluating every response for:

• Coherence — Internal consistency of the output
• Grounding — Whether claims are supported by the context provided by UCE
• Alignment — Whether the output addresses the original user intent

Configure minimum thresholds per agent or per workflow. Outputs below threshold are flagged for human review, sent back for revision, or blocked entirely.

How to Enable

NERO is opt-in during alpha. Enable it in workspace settings and configure your thresholds. We recommend starting with permissive thresholds and tightening based on observed output quality.

Performance

NERO adds 50-200ms of latency depending on output length. For most use cases, this is imperceptible and a worthwhile tradeoff for output quality assurance.

Feedback during alpha will shape the GA release. Report issues and suggestions through the dashboard feedback widget.

Audit Logging

Comprehensive audit logging is now available across all tiers. Every agent action, configuration change, user access event, and system operation is captured with full provenance. Export logs in JSON or CSV for SIEM integration.

Retention: Explorer (24 hours), Builder (30 days), Partner (90 days), Enterprise (configurable up to 7 years).

TypeScript SDK v2

The AEOS TypeScript SDK has been updated with:

• Improved type inference for agent skill definitions
• New AgentContext utility types for UCE integration
• Streaming response support for long-running agent workflows
• Better error types with structured error codes

Agent Disable

You can now disable an agent without deleting it. Disabled agents retain their configuration, skill manifests, and execution history but don't accept new tasks from MOSAIC. Re-enable with one click when you're ready. Useful for maintenance windows and gradual rollouts.

We've refined our pricing based on six months of usage data:

Updated Tiers

• Explorer (Free) — Increased from 100 to 250 agent actions/day
• Builder ($249/mo) — Now includes 25,000 agent actions/month (up from 10,000)
• Partner ($999/mo) — Now includes 150,000 agent actions/month with priority AWE runtime
• Enterprise — Custom quotas with dedicated infrastructure options

Usage Calculator

The pricing page now includes an interactive calculator. Input your expected agent count, average actions per agent, and orchestration complexity to get a cost estimate. The calculator uses real performance data from production workloads to provide accurate projections.

Grandfathering

Existing workspaces on paid tiers are grandfathered at their current pricing for 12 months. The new quotas apply immediately — you get more capacity at the same price.

Performance

• UCE context retrieval P95 latency reduced from 320ms to 180ms through query optimization and connection pooling
• Dashboard API responses are now gzip-compressed, reducing payload size by 70%
• Agent deployment time reduced by 50% through parallel dependency resolution

Analytics

• New filtering options: filter by agent, workflow, time range, and action type simultaneously
• Added funnel visualization for multi-step workflows showing drop-off at each stage
• Export now supports scheduled reports delivered via email or webhook

Billing Page

The billing page has been redesigned with clear visibility into:

• Current tier usage vs. limits for agent actions, context sessions, and orchestration calls
• Projected monthly cost based on current usage trajectory
• Detailed per-agent and per-workflow cost attribution
• One-click tier upgrade with immediate effect