Argus
VIEW
GITHUB
NeverForgetaPromiseAgain.

AI-powered WhatsApp memory assistant. Surfaces forgotten promises and recommendations exactly when you need them.

Powered by

Llama 3.3 70B + DigitalOcean Gradient

ARGUS

Argus is the AI-powered memory layer for WhatsApp. It uses Llama 3.3 70B via DigitalOcean Gradient to surface forgotten recommendations, commitments and promises in under 500ms — exactly when you need them.

<500ms

Ingestion Time

94.2%

AI Accuracy

<50ms

Context Check

1M+

Messages Scale

Architecture

Argus connects to WhatsApp via Evolution API, processes messages through Llama 3.3 70B on DigitalOcean Gradient for entity extraction, and indexes into SQLite FTS5. A Chrome extension monitors your browser and surfaces memories proactively. An ADK LangGraph agent handles conversational queries.

Technology Stack

Llama 3.3 70BDigitalOcean GradientGradient ADKNode.js BackendChrome ExtensionSQLite FTS5LangGraph AgentReal-time Webhooks
Features

Gradient Serverless Inference powers every AI function in Argus — event extraction, conflict detection, popup generation, relevance validation and natural language chat. OpenAI-compatible endpoint at inference.do-ai.run with automatic fallback to regex heuristics (Tier 2) or LRU cache (Tier 3) if Gradient is unavailable.

Llama 3.3 70BGradient ServerlessAI Fallback TiersJSON Repair

A standalone Python/LangGraph agent deployed via the Gradient ADK handles all AI Chat queries. It iteratively calls back into Argus through secured internal APIs — searching events, fetching details — then synthesises a natural language answer. Falls back to direct Llama call if the agent is unreachable.

Gradient ADKLangGraphPython 3.12Tool Calling

Real-time message ingestion via Evolution API. Every message captured, processed through Llama 3.3 70B for entity extraction and indexed into SQLite FTS5 in under 500ms. Supports individual chats, group conversations and media captions seamlessly.

Evolution APIWhatsApp BusinessWebhooks<500ms Latency

Proactive browser extension monitors your current URL and matches it against your indexed WhatsApp history using SQLite FTS5. Surfaces contextually relevant memories as subtle, non-intrusive popups — no searching required, ever. Eight distinct popup types for every scenario.

Chrome ExtensionURL Pattern MatchFTS5 Context EngineSmart Popup
Case Studies

01 / 04

TRAVEL INTELLIGENCE

Location Memory

Three months ago your friend recommended a cashew shop in Calangute. You forgot. Now you're booking a Goa trip — the moment you search 'Goa hotels', Argus surfaces that exact WhatsApp message. Shop name, location, confidence score. Zero effort.

<500ms

Ingestion

0.94

Confidence

VIEW ALL USE CASES

A — Z

Everything Argus does, from A to Z

AAI Memory Layer
BBM25 FTS5 Relevance
CContext Engine
DDigitalOcean Gradient
EEvent Extraction
FFuzzy Matching
GGradient Serverless
HHistory Search
IInstant Recall
JJust-In-Time Memory
KKnowledge Graph
LLangGraph Agent
MMessage Indexing
NNER Tagging
OOmnichannel Memory
PPromise Tracking
QQuery Speed <50ms
RReal-time Webhooks
SSQLite FTS5 Search
TTemporal Ranking
UUser Context
VVector Embeddings
WWhatsApp Native
XXact Confidence
YYour Data, Private
ZZero Search Needed

Download

Get Argus.

Available for all platforms.

Open Source

Star on
GitHub.

Argus is fully open source, submitted to the DigitalOcean Gradient Hackathon 2025. Built with Llama 3.3 70B via DigitalOcean Gradient Serverless Inference, Gradient ADK, and the Evolution WhatsApp API.

View on GitHub

Repository

ask-anannya/Argus-DO

License

MIT Open Source

Stack

Llama 3.3 70B · Gradient ADK · Node.js · Chrome

Hackathon

DigitalOcean Gradient Hackathon 2025