Ask your organisation. Get answers checked against your own sources.
ArgusBot answers staff questions in the chat they already use. It reads your code, databases and documents on your own network, says how sure it is, and stays quiet when it has nothing useful to add.
- Sources and index stay on your network
- Reads only, never writes
- No inbound ports
@ArgusBot how many deposits failed on Brand A yesterday, and why?
Yesterday Brand A had 1,284 deposit attempts and 97 failures (7.6%). Most came from one payment provider timing out between 21:00 and 22:10.
- provider timeout
- 61
- card declined
- 24
- limit reached
- 12
The player who raised it is j***@*****.com. Both of their attempts were timeouts.
The heavy part runs next to your data. The answering engine stays with us.
You run one small worker, the Edge, on a server you choose. It holds your credentials, copies and indexes your sources, and runs every lookup itself. It connects out to the ArgusBot cloud only to ask the model what to do next.
Your network
The Edge worker: one binary, or docker compose- Credentials, in a local vault whose key never leaves the host
- Copies of your sources and the search index built from them
- Every query and file read, under row, size and time limits
- Chat intake for Slack and Telegram, and posting the replies
- The full audit log of every question and answer
ArgusBot cloud
Hosted in the EU- The answering engine: model routing, instructions, the confidence check
- The confidentiality and personal-data check on every reply
- Knowledge it writes about your systems, kept per organisation
- The console: teams, channels, tiers, usage and the off switch
- The AI model keys, so you never need your own
A lookup arrives with limits such as rows=40 bytes=64000 seconds=20. The Edge applies the lower of those and your own ceiling, and refuses a lookup with no limits.
The Edge connects out. Slack runs in Socket Mode and Telegram by long-polling, so nothing on your side listens to the internet.
Answering stops. You keep your Edge, your data, your index and the channel connections. Knowledge we wrote stays exportable for 30 days.
Rules every answer keeps, on every channel
They come from an assistant that has answered support, finance and risk teams in production since August 2026. They are built in, not settings someone can forget.
Says how sure it is
Every answer ends with a confidence figure and the sources it checked, so people know when to double-check.
Quiet when it has nothing
If it cannot add anything useful, it posts nothing. No filler, no guessing out loud.
Masks emails and phones
Personal contact details are masked on every channel before a reply is posted.
Never shows a secret
Keys, tokens and passwords are caught and removed, even when a source contains them.
Scoped to the asker
A channel only sees the brands, data and people it is cleared for. Internal names stay internal.
Answers in your language
Ask in Spanish and get the answer in Spanish, from the same English sources.
Connect what your people already trust
Each connector only reads, carries its own access labels, and gets a dry run before it goes live.
Git mirrors, with maps of services, settings and event flows.
Postgres read-only, DuckDB and warehouse files, with schema digests.
HTTP and GraphQL endpoints on an allow-list you set.
Documents, with their existing sharing kept as access labels.
Wikis and runbooks, checked for staleness against the source.
Ticket systems, HR records with field-level masking, past threads.
Your IT team installs it without us
- Step 1
Create the Slack app
The console gives you a manifest. The app lives in your workspace, not ours.
- Step 2
Start the Edge
One binary with a systemd unit, or docker compose. No database to install.
- Step 3
Pair it
Enter the one-time pairing code. The Edge connects out and appears in the console.
- Step 4
Add sources
Type credentials into the Edge's own local page. They never pass through our cloud.
- Step 5
Map channels
Choose which channels can ask, their tier, and which brands they see.
Control it from one screen, not config files
Organisations and workspaces
Several Slack workspaces and Telegram groups under one organisation.
A tier per channel
Internal, operator and generic tiers decide what each channel may be told.
Audit and usage
Every answer with its confidence and cost, plus ratings from your staff.
Off switch
Pause one channel or the whole organisation in one click.
| Channel | Workspace | Tier | Brands | Language |
|---|---|---|---|---|
| #platform-eng | Company | internal | all | asker's |
| #finance-ops | Company | operator | Brand A, B | asker's |
| #support-brand-c | Partner | operator | Brand C | Spanish |
| Telegram: ops-night | Telegram | operator | Brand A | asker's |
| Direct messages | Company | generic | by roster | asker's |
Example channel map. Names are illustrative.
Try it on one channel, with your own sources
A 30-day pilot on one Slack workspace or Telegram group, set up with the team that builds it.