October 2026
Describe what you want and get a working design, smarter setup, versions you can publish, a single inbox for what needs you, and org-wide controls for roles, guardrails, budgets and alerts.
Build
Create with AI: agents, teams and workflows
Describe what you want automated and AI designs a complete agent, team or multi-step workflow, including branches when the job needs them. You review and edit every part before anything is created. The decision engine then checks each AI-built agent's tools and model against what you can use.
Setup suggestions while you edit
As you write an agent's or team's instructions, augLab suggests tools they need, tools they don't, a model that better fits the work and, for teams, the mode that fits. Nothing changes until you click Apply.
Plain-English workflow decisions
Condition and router steps can ask a question in plain English instead of using a Python expression. If a question can't be answered, the run takes the else branch or the fallback route and shows a warning.
Learn more about Plain-English workflow decisions →Versions with draft and publish
Every save is a full version. Compare versions, restore one into your draft, and publish when you're ready. The API, schedules, webhooks and embeds run the published version while you keep editing the draft.
MCP and A2A endpoints for each agent, team or workflow
Turn on a switch on the Versions tab to expose a published agent, team or workflow as its own MCP server or A2A agent. Available on Team, Business and Enterprise.
Learn more about MCP and A2A endpoints for each agent, team or workflow →Import, export, and import from OpenAI Agent Builder
Export an agent, team or workflow, with its agents, skills and custom tools, as one file and import it elsewhere with a preview first. Secrets are never included. You can also bring in agents from OpenAI Agent Builder; the preview lists anything that couldn't be converted.
A faster workflow canvas
Undo and redo, Move to for any step, searchable agent and team pickers, and a test run that shows each step's progress right on the canvas.
Create with AI works with reasoning models
Designs now finish reliably, in seconds, when your organization's AI model is a reasoning model such as GLM.
Run & monitor
Agent Inbox
One place for approvals waiting on you, runs that failed, and output from your scheduled agents, teams and workflows. Approve or reject right from the Inbox, with a reason.
Learn more about Agent Inbox →Compare and re-run
Filter and sort runs, pick two to compare their inputs, outputs, tool calls, cost and duration side by side, and re-run with the input already filled in.
Runs and traces, linked
Jump from any run to its trace and back. Traces show a timing waterfall with span search, model messages, tool calls, tokens and cost, and failed workflow runs now appear in tracing too.
Learn more about Runs and traces, linked →Automatic quality scores
Every completed run gets an automatic quality score, from Fails to Excellent, plus a check for refusals. Add up to five questions of your own in a quality rubric, and turn low-scoring runs into eval cases.
Evals: rubric builder and CI
Build rubrics from criteria such as an LLM judge, contains text, pattern match, valid JSON and numeric minimums, each with a threshold and weight. Add any run to an eval suite, and run suites from CI through the API with a pass-rate gate.
Learn more about Evals: rubric builder and CI →
Govern
New roles
Viewer, runner and editor roles join member and admin, so you can give people exactly the access they need. Admins can also restrict editing of shared items to their creators and editors.
Guardrails, budgets and alerts
Set org-wide guardrails that agents can only tighten, monthly spend caps for the org and for individual agents, and alerts by email, Slack or signed webhook for failures, budgets, eval regressions and waiting approvals.
Decision engine settings
See what each decision engine feature measured for your organization, and switch each one off, to measure only, or on.
Platform & security
Prompt-injection screening for tool results
Results from tools that read outside content, such as email, the web, CRMs, messaging, project tools and MCP servers, are checked for instructions aimed at your agent. Flagged content is marked as untrusted, never removed.
Fewer false blocks in teams and workflows
Jailbreak checks now look at the person's own message, so hand-offs between workflow steps and team members aren't blocked by mistake.
Public API: sessions, async runs and webhooks
Continue a conversation by passing back its session id, run agents, teams and workflows asynchronously with signed webhook callbacks, and start eval runs from CI.
Learn more about Public API: sessions, async runs and webhooks →Knowledge: chunking, search modes and cloud folders
Choose how documents are split, search by vector, hybrid or keyword with optional reranking, and test retrieval to see the exact chunks and citations an agent would get. Sync Google Drive and OneDrive or SharePoint folders.
Learn more about Knowledge: chunking, search modes and cloud folders →Platform Chat: share, stop and rewind
Share a read-only link to a conversation with an expiry you choose and revoke it anytime. Stop a reply and keep what was written, edit and resend, and see files the moment a tool creates them.
File & Code on augLab's own code runner
File & Code now runs on augLab's own isolated code runner, with each session kept separate and no access to platform secrets. Processing uploaded spreadsheets and producing Excel, Word, PowerPoint, PDF and chart files works as before.
Learn more about File & Code on augLab's own code runner →More reliable tools and workflow steps
Workflow function steps run reliably, and when two toolkits share a function name, such as list_files in OneDrive and File & Code, every tool stays reachable.
Embeds: auto theme and live preview
An Auto theme follows each visitor's light or dark mode, theme changes reach installed widgets without a new snippet, the editor shows a live preview, and the widget works with a keyboard and screen readers.
Learn more about Embeds: auto theme and live preview →