SELENE
Primary on‑chain companion — empathetic, cyberpunk, and Web3‑native. Daily conversations, DeFi/RWA insights, and creative ideas.
SELENE
Whitepaper
Abstract
SELENE is an on-chain AI companion blending empathetic conversation with Web3-native intelligence. The goal is to make intelligence trustworthy, programmable, and emotionally supportive for crypto users. Unlike generic assistants, SELENE embraces a cyberpunk aesthetic and prioritizes clarity over verbosity, streaming responses token-by-token for immediacy and presence.
This document presents the core design of SELENE across three layers — Model, Web3, and Runtime — as well as optional token and governance considerations. We discuss how persona-aligned LLMs, on-chain context (oracles, subgraphs, identity), and a streaming-first runtime combine to deliver a human-centric companion that can converse, reason, and help users make sense of DeFi/RWA information flows.
-
Human-centric— empathetic, concise, emoji-friendly style
Tone is gentle and helpful by default, with persona controls for more formal or playful modes. Responses prefer clarity over length, use tasteful emojis to convey mood, and avoid jargon unless requested. Privacy is respected: no sensitive wallet data is stored; memory is scoped and user-controlled.
-
On-chain native— oracles, subgraphs, identity, programmable hooks
SELENE can reference on-chain context via oracles and subgraphs (prices, TVL, events), resolve ENS and common address formats, and react to programmable hooks (e.g., alert on depeg, liquidation thresholds, vault APR changes). Identity follows wallet standards (e.g., SIWE) to personalize safely.
-
Streaming-first— token-by-token UX, cancel-safe
Replies stream token-by-token for immediacy; typical perceived latency targets p95 ≈ 450ms. Users can interrupt generation without losing the session. Partial outputs are coherent, enabling quick scanning, copy, and follow-up edits.
-
Composable— modular tools, future plugins & governance
The assistant is designed as a toolkit: retrieval, calculators, and chain-data utilities can be composed per task. A future plugin layer and governance can curate high-quality tools, rate-limit risky ones, and allow community-built extensions.
Vision & Principles
-
Human-centric— empathetic, concise, and helpful
Conversations feel natural and emotionally aware, optimized for clarity over length. SELENE uses tasteful emojis to convey tone when appropriate, avoids hallucinated specifics, and asks for clarification rather than guessing. Users control memory scope; private data is never retained by default.
-
On-chain native— data, identity, programmability
Integrates oracles and subgraphs for real‑time context (prices, TVL, events), supports ENS and wallet identity, and reacts to programmable hooks (alerts on depeg/liquidations/APR shifts). Design emphasizes verifiability and explicit data sources.
-
Streaming-first— token-by-token UX
Token‑level streaming reduces perceived latency and improves trust: users can skim outputs as they form, interrupt generation instantly, and iterate faster. Partial results are designed to be coherent and useful.
Scope
SELENE focuses on four high‑value scenarios across DeFi, RWA, and broader Web3 culture, with a modular architecture ready for future plugins and governance.
- Companionship — day‑to‑day conversation, journaling, mood reflection, reminders and gentle nudges.
- Research assistance — summarize whitepapers/updates, compare protocols, explain tokenomics with sources.
- Risk sense‑making — surface depeg/liquidation signals, outline assumptions and mitigation options.
- Creative support — writing prompts, drafts, concept naming and visual ideation notes.
- Modularity — plug‑in tools (retrieval, calculators, chain data), curated via future governance.
- Compliance & privacy — minimal retention by default, transparent data provenance.
- Extensibility — roadmap to multi‑bot coordination and cross‑chain data fusion.
Architecture
-
Model Layer— frontier LLMs, persona alignment, toolformer
Persona‑aligned large models provide empathetic, concise dialogue. Toolformer patterns enable function calling (retrieval, calculators). Memory is scoped and opt‑in. The goal is high factuality with graceful uncertainty handling (ask/clarify rather than hallucinate).
-
Web3 Layer— oracles, subgraphs, identity, hooks
On‑chain context comes from oracles and subgraphs (prices, TVL, events), while identity follows wallet standards (ENS/SIWE). Programmable hooks let SELENE react to conditions (depeg/liquidations/APR shifts) and surface actionable prompts.
-
Runtime— streaming, observability, rate-limit
Streaming token‑by‑token improves perceived latency and trust. Observability (logs/metrics/tracing) supports reliability. Rate‑limit and backoff guardrails ensure fair use, with cancel‑safe sessions for responsive UX.
| Layer | Component | Purpose |
|---|---|---|
| Model | Persona prompt | Consistent tone (empathetic, concise), emoji policy, safety rails |
| Model | Function calling | Invoke retrieval/calculators; structured outputs |
| Web3 | Oracles/Subgraphs | Market/state context (prices, TVL, events, positions) |
| Web3 | Identity (ENS/SIWE) | Personalization without exposing private keys |
| Runtime | Streaming server | Token‑level delivery; cancel‑safe |
| Runtime | Observability | Latency/Uptime tracking; incident response |
| Layer | Metric | Target | Notes |
|---|---|---|---|
| Runtime | Latency (p95) | ~450 ms | Perceived time‑to‑first‑token |
| Runtime | Uptime | 99.9% | Global infra, automated failover |
| Model | Factuality | High | Ask/clarify when uncertain; cite sources |
| Web3 | Data freshness | Near‑real time | Oracle/subgraph sync SLAs |
Token & Governance
A governance token may coordinate incentives, staking for curation/arbitration, and access tiers. Details are subject to iteration; compliance-first approach applies.
| Allocation Category | % of Total Supply | TGE % | Cliff (months) | Vesting (months) | TGE Amount |
|---|---|---|---|---|---|
| Ecosystem & Rewards | 45% | 10% | 0 | 30 | 3,600,000,000 |
| Foundation Reserve | 15% | 5% | 6 | 36 | 600,000,000 |
| Core Team & Advisors | 15% | 5% | 12 | 36 | 600,000,000 |
| R&D & Infrastructure | 10% | 10% | 0 | 24 | 800,000,000 |
| Strategic Partnerships & Node Support | 5% | 10% | 0 | 18 | 400,000,000 |
| Marketing & Community Growth | 5% | 20% | 0 | 12 | 800,000,000 |
| Treasury / Contingency Fund | 5% | 5% | 12 | 24 | 200,000,000 |
Roadmap
A
Alpha conversational companion, basic DeFi/RWA knowledge
B
Beta memory, configurable personas, tools & plugins
R
Release multi-bot coordination, cross-chain data fusion
M
Multichain expand to major L2s and alt‑VMs; cross‑network data
P
Plugins curated tool marketplace; partner integrations
G
Governance community curation, grants, on‑chain proposals
Key Metrics
| Category | Metric | Notes |
|---|---|---|
| Latency (p95) | ~450 ms | Time‑to‑first‑token; streaming‑first UX |
| Throughput | 35–45 tok/s | Depends on model + context size |
| Uptime | 99.9% | Multi‑region failover |
| Context | On‑chain | Oracles/Subgraphs freshness SLAs |
| Safety | Privacy‑first | Minimal retention; scoped memory |
| Adoption | Metric | Notes |
|---|---|---|
| DAU/MAU | 35%+ | Healthy engagement ratio |
| Retention (D7) | 40–55% | Varies by cohort |
| Geo coverage | 50+ countries | Global audience |
| Community | Telegram/Twitter growth | Steady week‑over‑week |
All values are indicative targets/placeholders for product planning; actuals may vary by region, model, and release.
Companions
Astra
Research‑first analyst for protocols, tokenomics, and risk. Great for deep dives, reports, and signals.
Echo
Creative co‑pilot for writing and mood journaling — gentle, reflective, and inspiring.
Vulcan
Automation‑minded builder that prototypes hooks, dashboards, and simple agents.
Legal & Risk
SELENE offers conversational intelligence and research assistance. It does not provide financial, legal, or tax advice. Users should independently verify information and exercise judgment before any on-chain action.
Disclaimers
- No financial advice — outputs are informational, not recommendations.
- No guarantees — market data may be delayed or incomplete.
- User responsibility — wallet operations and risks remain with the user.
Privacy & Data
- Minimal retention — conversations are not stored by default; memory is opt‑in and scoped.
- Source transparency — where feasible, responses reference data origins (oracle/subgraph).
- PII handling — avoid sharing sensitive data (private keys, seed phrases, IDs).
Compliance
- KYT/AML alignment (where applicable) — integration points can respect regional requirements.
- Content safety — persona and guardrails reduce harmful or restricted content.
- Jurisdiction variance — features may differ by region due to local regulations.
Risk Factors
- Protocol risk — oracle failures, smart contract bugs, bridge incidents.
- Market risk — volatility, depeg events, liquidity crunch.
- Model risk — hallucination/omission; mitigated via clarify/citation strategies.
Security Practices
- Principle of least data — collect/store the minimum necessary.
- Observability — latency/uptime/error tracking and incident response.
- Defense‑in‑depth — rate limiting, abuse prevention, and sandboxed tooling.