Agents, everywhere you already work
This is the most ambitious release AANCER has shipped to date: six major programmes, delivered together, independently penetration tested with zero open findings at cut. Chat stops being a text box and starts acting: mention any connector, workflow or agent with @ and watch it run, with a confirmation card standing guard before anything writes, sends or deletes, and up to 100 actions behind a single approval. Agents graduate into a governed, versioned product with their own editor, evaluations and analytics. The assistant remembers you, works inside shared folders, and carries a skills library your team can make mandatory. Chats become projects, share links, and recurring scheduled tasks that run themselves. For teams ready to open the doors further, the same governed assistant now answers inside Microsoft Teams and Slack, and a published agent can be called like any model, over MCP, or by another organisation's own agent entirely, over A2A. Spend caps, automatic compliance checks before publish, and one inventory tracking every agent's risk class, checked across 540 tested combinations of surface and outcome, close the loop for governance teams. Six programmes. One release.
- Chat that actually does things
Type @ and bring in any connector, workflow or agent you're allowed to use. Describe what you want in plain language and the platform fills in the fields itself. Anything that writes, sends, updates or deletes stops for an editable confirmation card first, so you stay in control of every action, not just every answer. Need to do this dozens of times over? One approval runs the whole batch, up to 100 items, with per item progress.
- Agents, built like real software
Every agent now lives in its own editor: configure it, test it, let AI build or refine it, compare versions, run evaluations against real cases, and watch its usage and cost in one analytics view. Publishing locks a version to its exact configuration, so nothing changes under a workflow's feet without you choosing it. An AI Router step reads what a message means and sends it down the right branch automatically.
- An assistant that remembers, and a library it can work in
Personal memory keeps what matters to you, and only you; sensitive categories like health or religion are refused outright, by design. Shared library folders hold versioned files your chats can read and write directly. Skills teach the assistant a repeatable procedure once, so your team never has to re-explain it, and admins can make the ones that matter mandatory.
- From one-off chats to running operations
Projects group chats, instructions, a shared folder and your favourite agents under one roof. Any chat becomes a read-only link you control, revocable any time. And the request that used to mean a recurring calendar reminder, “every Monday at 8, summarise...”, is now a scheduled task the assistant runs for you, every time, reporting straight to your inbox.
- Your agents, wherever your team already is
Mention the assistant inside a Microsoft Teams or Slack conversation and it answers with exactly the access and audit trail it would have on the web, nothing more. A published agent becomes a model any OpenAI-compatible tool can call, a capability exposed over MCP, or a public agent card another organisation's own agent can reach over A2A. One assistant, reachable everywhere your work already happens.
Lizenzgebunden · off by default; turned on per licence (chatSurfaces) - Governance that keeps up with the agents it's watching
Spend caps apply per workspace, per user, per agent or per channel, and a cap that's reached can fall back to a local model or stop outright, your call. Compliance rule packs check every agent automatically before it can be published, not after. And one inventory lists every agent in the business with its owner, its AI Act risk class and an automatically computed operational risk score, so governance teams see the whole fleet at a glance.
Lizenzgebunden · Professional and Enterprise editions (governanceRules) - Code steps get real dependencies, safely
Approved libraries are now usable the moment they're approved, not after the next release. A dedicated curator fetches only from the registry an administrator names, with every step audited, so custom code in your workflows gets the packages it needs without opening the platform to the open internet.
- Service-account API keys, built in
Give an integration its own key, with its own budget, its own rate limit and its own model restrictions, no third-party key broker required. Code steps can now read secrets by name from a dedicated, encrypted secrets store, and an automation can be triggered directly by an API key instead of a person.