← Back to all sparks
G

Guru

COLLAB
Velocity5.0

AI-powered enterprise knowledge management, internal wiki, and knowledge-agent platform

Guru is tightening the leash on the Knowledge Agent it just handed the keys to.

knowledge-agentsslackagent-groundingguardrailspermissionsautonomy
Current state
Everything recent is the Knowledge Agent. It already writes to the knowledge base — creating collections, bulk-archiving cards, running drafts through to publish — and the newest work narrows where and how it acts: per-channel behaviour in Slack, and skills pinned to a named document instead of searching for one. Around that sit guardrails, skill permissions, recoverable drafts and scheduled automations.
Where it's heading
The arc is autonomy first, then control. Having given the agent write access and a schedule, Guru is now spending releases on determinism and scoping — making the agent's behaviour predictable per channel and its grounding exact per skill. That is the work required before customers let an agent run unsupervised on a knowledge base, and it suggests trust, not answer quality, is the constraint being engineered against.
Prediction
The scoping controls point toward auditability as the next gap: some record of what the agent did to the knowledge base, and per-skill or per-channel review of those actions, to match the permissions and guardrails already shipped.

Recent moves

  1. 14d ago

    Tune how your Slack agent joins the conversation, channel by channel

    Slack Knowledge Agents get per-channel response settings, replacing one behaviour everywhere. It is a scoping control of the same kind the recent releases keep adding: the agent's reach is unchanged, its predictability in a given room is not.

    View source ↗
  2. 14d ago

    Ground a skill in the exact document it should use, every time

    Skill References attach a specific document to a skill so it stops relying on search to find the right file. Grounding moves from instruction-and-hope to a fixed pointer, which is the determinism half of the trust work.

    View source ↗
  3. 1mo ago

    Hand your knowledge base upkeep to your Knowledge Agent

    ⚡ SPARK

    This is the release the newer scoping controls exist to contain: the agent gained write access to the knowledge base, from bulk archiving to the full draft-to-publish loop. Everything since has been about bounding what that autonomy can touch.

    View source ↗
  4. 1mo ago

    Never lose a draft to employee turnover again

    Drafts now survive offboarding and admins can see and recover every draft in the workspace. A recoverability fix that matters more once an agent is also producing and publishing drafts.

    View source ↗
  5. 1mo ago

    Operator Mode

    Operator Mode lets admins configure the Knowledge Agent by describing what they want instead of walking settings screens. It lowers the cost of tuning an agent whose behaviour the later per-channel and per-skill controls make finer-grained.

    View source ↗
  6. 2mo ago

    Microsoft Work IQ

    Knowledge Agents connect to Microsoft Work IQ, reaching Outlook, SharePoint, OneDrive, Teams and Office files in real time. Reach into the Microsoft estate broadens the agent's inputs without changing what it is allowed to do.

    View source ↗