One workspace for every AI answer, agent and citation
The home for AI in the cloud edition — threaded conversations with a copilot switcher, a prompt library, an insight feed you triage, agents that must be approved before they act, a knowledge hub that cites its sources, and a governance page that meters the lot.
A workspace, not a chat box
A searchable thread rail you can pin and archive, a centre pane that switches copilot per question, and a right-hand feed of scored insights you dismiss or mark actioned. Attach a document to a thread and ask questions of it in place. Every answer shows the grounding it was handed — the underlying data is always one “Data” toggle away.
The right copilot picked deterministically
A pure router reads the question and decides which copilot and which data lenses to gather — finance, sales, procurement, budget, and the module copilots for service, support, contracts, tenders, subscriptions, manufacturing, distribution, retail, construction and real estate. Lenses belonging to packs you haven’t enabled are inert by construction, so a disabled module can’t leak into an answer.
Agents that ask before they act
An agent plans against a whitelisted tool registry — create a follow-up task, draft a campaign, log a problem record, open a service job card, raise a milestone invoice — and any risky plan parks at awaiting approval. Approve and execute runs only existing, audited mutations through the same permission checks staff use; reject is terminal with zero side effects. The planner itself is deterministic, so it still drafts a plan with AI switched off.
A knowledge hub that cites
Ingest text and it is deduplicated by hash and chunked on sentence boundaries server-side. Retrieval is always deterministic keyword ranking with citations back to the exact chunk; where embeddings exist, cosine similarity blends in. An LLM answer layers on top only when narration is enabled — the passages you are shown don’t change either way.
Governance you can show finance
Per-feature usage — calls, prompt and completion volume, average latency and how many were actually served by AI — plus day-idempotent background schedules and a gate card that states plainly what still works with AI turned off.
What’s included
- Threaded conversations — search, pin, archive, start fresh
- Copilot switcher with deterministic intent routing across enabled packs
- Prompt library with categories, favourites and fill-in variables
- Chat attachments and in-thread document Q&A
- Scored insight feed with dismiss / actioned triage
- The grounding snapshot visible on every answer
- Agents over a whitelisted tool registry, human approval before execution
- Argument validation that refuses out-of-scope or malformed tool calls
- Knowledge hub — hash-deduped ingest, chunking, cited retrieval
- Per-feature usage metering, background schedules and an AI-off gate card
Frequently asked
How is this different from the AI Copilot and AI CFO?
Those are answers — a conversational copilot and a board-style CFO briefing. The workspace is where they live: thread history, the prompt library, the insight feed, the agents, the knowledge hub, and the governance page that meters every call.
What still works with AI turned off?
Copilot routing, insight scoring, the agent planner and knowledge-hub retrieval are all deterministic and need no model. What needs connectivity is the written narration layered on top. The governance page carries a gate card that spells this out per feature.
Can an agent do something I didn’t sanction?
No. Each agent has an explicit allow-list of tools, and a plan step naming an unknown tool, a tool outside that list, or malformed arguments is refused before anything touches the database. Execution calls the same audited mutations and permission checks as the UI — nothing bypasses them — and anything risky waits on a human approval.
Where do the AI keys live, and what does it cost?
On our gateway, never on your machine, and usage is metered — the workspace shows it per feature. The conversational AI itself is the Enterprise add-on; the deterministic routing, retrieval, planning and scoring described here are not.
What exactly is included in AI Workspace · Cloud edition?
threaded conversations — search, pin, archive, start fresh, copilot switcher with deterministic intent routing across enabled packs, prompt library with categories, favourites and fill-in variables, chat attachments and in-thread document Q&A, scored insight feed with dismiss / actioned triage and the grounding snapshot visible on every answer, and 4 more. Every item on that list is a working screen you can open in the trial rather than a capability described in the abstract.
How is ai workspace · cloud edition handled differently here?
It rests on a workspace, not a chat box, the right copilot picked deterministically, agents that ask before they act, a knowledge hub that cites and governance you can show finance. The distinction that matters is that these are structural rather than cosmetic — the behaviour is built into how records are posted, not layered on as a report you have to remember to run.
Does AI Workspace · Cloud edition work offline?
On the Desktop edition, yes — completely. threaded conversations — search, pin, archive, start fresh and copilot switcher with deterministic intent routing across enabled packs run against a local database on your own machine, so the screens behave the same with the network unplugged as with it connected. AI and billing are the only two things that reach out. On the Cloud edition ai workspace · cloud edition runs in the browser and needs a connection.
Does AI Workspace · Cloud edition post to the real books?
Yes, and that is the whole reason it lives in this product rather than beside it. What a workspace, not a chat box records lands in the same posted double-entry ledger the Trial Balance, P&L and GST returns are built from — so there is no second set of ai workspace · cloud edition numbers to reconcile against the first.
How does AI Workspace · Cloud edition affect my GST?
It feeds it directly rather than sitting alongside it. CGST/SGST and IGST are split from HSN or SAC and place of supply as each document is raised, so whatever ai workspace · cloud edition produces is already correct when GSTR-1, 3B and 2B reconciliation are drawn off the same books. The return matches the accounts because it was never a separate exercise.
Can I try AI Workspace · Cloud edition before committing?
Yes — 14 days, no card. Test ai workspace · cloud edition against your own masters and your own transactions rather than sample data: the questions worth answering here are about your business's edge cases, and a demo dataset is built not to have any.