Bugis.ai uses secure, locally hosted AI to detect viral misinformation early, draft factual counter-narratives, and ensure total Whole-of-Government alignment — so every ministry and statutory board is pushing the same verified information to citizens in real time, safely and securely.
Built for governments, ministries, and operators of critical systems
The product
From the first signal to the post-crisis report, every step is tenant-scoped, classification-aware, and tamper-evident.
DETECT
Ingest signals from surface web, social, broadcast, dark web, and messaging. Score them with named factors. Cluster into narratives. Route to the right operator.
RESPOND
A morning brief grounded in your approved fact-base — with explicit fact-gaps, never a fabricated line. Drafts move through approval chains and dispatch with operator confirmation.
LEARN
Auto-generate post-crisis reports, attribute coordinated campaigns, propose playbook updates, and rehearse the next event in a clearly-marked simulation sandbox.
Configurable + extensible
Bugis ships with a working baseline at every layer of the stack. Your tenant tunes what's already there; we build the bespoke pieces — adapters, factors, channels, agent tools — when your sector calls for them.
Inputs
Source adapters
RSS · GDELT · Reddit · HN · YouTube · Telegram · Tor
Processing
5-factor scoring
Credibility · Watchwords · Recency · Reach · Intensity
Core
Bugis platform
Detect · Respond · Learn · Audit
Agents
Tool-using surfaces
Investigation Agent · Ask Bugis
Outputs
Channels
Press · Social · Internal · Broadcast
Runtime
LLM & sovereign infra
Cloud LLM · Local LLM · National LLM · OpenBao · k3s
Extension points
Add a custom RSS source, scrape a dark-web forum behind Tor, ingest a national broadcaster's API, or pipe in your SOC's existing intel feed.
The five-factor composite is the default. Add tenant-specific factors — election integrity, supply-chain disruption, brand-safety bands — with their own weights.
Plug in your own approval chain, classification ladder, or dispatch channel. Trigger.dev workflows are just code, version-controlled per tenant.
Extend Ask Bugis with scoped tenant tools — query your CRM, fetch a regulator's docket, kick off a war-room invite. Read freely; writes always confirm first.
One abstraction, any backend. Anthropic for cloud, Gemma 4 for sovereign on-prem, or any OpenAI-compatible national model. Routing is policy-driven.
Sector-specific surfaces — election rapid-response, telco outage war-room, central-bank rumour desk — built alongside the core, not bolted on.
Determinism, not vibes
Every signal carries five named factors and a composite. Weights are tenant-tunable and stored. The same input produces the same score on every recompute — and the model renders an explanation an operator can read in five seconds.
tenant_scoring_config— every change writes an audit row.composite score
82
f_source_credibility
f_watchword_relevance
f_recency
f_reach_potential
f_emotional_intensity
Same input · same weights · same score. Always.
Built for sovereignty
The same engineering rules whether you run on Anthropic in the cloud or a sovereign on-prem cluster with a national LLM behind the firewall.
Every line the assistant generates cites an approved fact. Missing facts surface as gaps for an operator to resolve — never invented to fill the page.
Cloud or local LLM behind the same abstraction. Sovereign deployments pin to a customer-approved model; the UI banners which mode is in use.
Postgres row-level security enforces tenant scope on every query — including admin tools. Cross-tenant visibility flows only through Central Command.
Every state-changing action writes a hash-chained audit row: actor, role, action, target, before/after, source IP, classification, token cost when an LLM is involved.
Agentic surfaces
Tool-using agents that read freely, write only with operator confirmation, and leave a full trace behind.
Pinned to a single signal. Pulls related signals, fetches the source URL, calls structured tools, and returns claims as supported / unsupported / contested with a recommendation.
tools
Page-aware floating assistant with scoped tenant tools. Reads freely; writes only after a dry-run preview and operator confirmation. Doubles as the agentic onboarding flow.
tools
How we work with you
Bugis is a productized platform with a consultative motion. You don't buy a SaaS and hope it fits — we configure and extend the platform to fit, on a four-phase cadence.
We map your sector's narrative surface area: where stories form, who shapes them, what your team needs to see first.
We tune the baseline — sources, watchwords, scoring weights, classification ladder, approval chain, IdP — to your tenant.
Bespoke pieces — custom adapters, sector-specific scoring, agent tools that talk to your existing systems — built alongside the core.
Sovereign-hosted or managed; SLA backed; with a learning loop that turns every crisis into a playbook update.
Built around a documented capability spec — so every feature ID maps to a story, a test, and a row in the audit log.
Built around roles
Lives in the operator console. Triages signals, opens investigations, drafts the morning brief.
Owns sources, watchwords, lawful-monitoring policy, and the team. Signs off on dispatches.
Oversight across tenants in their jurisdiction. Reads everything; never acts inside a tenant.
Runs the platform. Owns secrets, audit, IdP bindings, and the tenant lifecycle.