OneByZero Neo 6.0 was released on 30 June 2026. This release makes AI Coworkers generally available and adds Skills, Neo Code Agents, and human approval checkpoints across every layer of agent execution.
What’s New
AI Coworkers — generally available
- Ready for production — Coworkers can now be deployed across the enterprise, with governance controls, channel integrations, skills, and long-term memory.
- Microsoft Teams and Outlook — Coworkers send and receive messages on Teams and reply to emails in Outlook, with the full reply thread as context.
- Feedback from Teams reactions — a thumbs-up or thumbs-down on a Teams message is captured as feedback and shapes the Coworker’s next replies.
- Long-term memory — each Coworker keeps its own persistent memory in its workspace, with a status badge showing whether memory is active. Reflection on Teams reactions has its own toggle.
- Conversation-end feedback — a prompt at the end of each conversation feeds into the Coworker’s reflection.
- Guardrails — attach guardrails when you create or edit a Coworker to screen every conversation against your content policies.
- Analytics and KPIs — a dedicated analytics view shows usage, conversation volumes, and trends for each Coworker, and the KPIs tab shows a card for each metric. KPI definitions carry over when you publish a new version.
- Redesigned builder — a faster setup wizard, breadcrumb navigation, search and filters on the Coworkers list, and version history for users with view-only access.
- Conversation actions — mark conversations complete or delete them from the conversation list.
Skills and Skillsets
- Skills — package instructions, tools, and scripts as reusable skills, written in Markdown and managed per workspace.
- Expose as Skill — turn any agent or Coworker into a reusable skill with one toggle.
- Skillsets — group related skills into named Skillsets under Integrations, each with its own access policies.
- Approval for skills — a skill can require human approval before it runs.
Neo Code Agents (Beta)
- Connect externally hosted agents — register an agent that runs in your own service and manage it alongside native agents. It appears as Remote Agent on the Create page and as its own category on the dashboard.
- Same channels and access controls — Neo Code Agents connect to channels and follow the same roles and access policies as native agents.
- Off by default; your platform administrator turns it on.
Human in the Loop
- Approvals everywhere — pause for human review before a tool call, an MCP tool call, or a skill runs.
- Several approvals in one conversation — an agent can pause and resume more than once in the same session.
- Business-rule approvals — on a Coworker, describe the situations that need approval and who approves them.
Agentic Flow
- Break and Resume — pause a flow mid-run and continue when an external system sends a signal. Several pauses can be chained in one run.
- File Type Identifier — detect an uploaded file’s type and route it accordingly.
- ZIP files — flows accept ZIP uploads as attachments.
- Larger files — uploads of up to 500 MB.
- API Request — OAuth2 authentication, request bodies built with the Text Template component, and non-2xx responses reported as flow errors.
- Configurable agent timeout — the fixed 30-second limit on agent steps is gone.
- Output node — more flexible delivery routing, and Adaptive Cards for richer replies in Teams.
- Upload and run in one call — upload files and start a flow with a single API call.
- Faster runs — independent paths in a flow run in parallel.
LLM models
- One place for models — configure a model once and it is available across the platform.
- External provider — connect privately hosted model endpoints.
- Amazon Bedrock — all Bedrock models are grouped under one Amazon provider.
- Model descriptions — each model shows a plain-language description to help you choose.
Knowledge bases
- API-key authentication for Bedrock knowledge bases.
- Paginated document lists and faster loading of the knowledge base list.
- Configurable search connection timeout.
Voice agents
- New speech-to-text providers — OpenAI GPT-Realtime and Soniox.
- Voice on your own infrastructure — a new voice service for organisations that run voice on-premises.
- Cost tracking — speech-to-text and text-to-speech usage is included in traces and cost reporting.
- Custom LLM providers — voice agents can use custom LLM providers.
Access control
- Consistent enforcement — roles and access policies apply across agents, flows, Coworkers, skills, and channel credentials.
- Policies at invite — attach access policies when you invite members or import them in bulk.
- Channel credential policies — control which users and roles can use each channel credential.
- Sharing Coworkers — share a Coworker with colleagues.
- Clear permission messages — users who try an action outside their role see an explanation.
- Policy management — delete several policies at once; duplicate policy names are blocked; the policy list is paginated.
Observability
- Full LLM tracing — traces cover every service that uses a model, including voice, knowledge base queries, and skills.
- Cost attribution — costs are captured and shown in trace records.
- Component logs — execution logs for each flow component.
Changes
- The traces list no longer shows raw input and output columns. Open a trace to see them.
- Skillsets moved from the Coworker builder to the Integrations page.
- Flow names and endpoint paths are scoped to the workspace, so flows in different workspaces no longer conflict.
- Contact directories, user directories, knowledge base documents, and the Skills tab are paginated.
Fixes
- Structured output returns only the fields defined in the output schema.
- Files attached with the File component upload correctly.
- Document parsing no longer drops content from chunks.
- Streaming works with an output guardrail enabled.
- Every Coworker no longer shows the same KPI data.
- Workspace sessions no longer expire during long working sessions, and switching workspaces redirects correctly.
- Security vulnerabilities reported in platform dependencies are resolved.