your sources Pulse verification named agents guarded execution

An intelligence layer over the sources you actually trust.

Connect your own sources to a verified Pulse live feed and deploy named agents with separate risk knobs. Fenestra covers crypto perps and memecoin calls, plus research for equities, political markets and options. Every guarded execution path is approve-first by default.

Closed pilot · small waves · email used only for access updates.

Approve-first by default
illustrative
SOL-USDT · perp LONG
Entry zone141.2–142.6
Stop138.9
Target 1147.8
R / R2.4

source tg / momentum-desk · score 0.72 · 38 tracked calls · sizing ×0.8

"Reclaim of range low with funding reset; invalidation is a close below 139." — extracted thesis, evidence linked
TP/SL atomic · caps enforced
kill switch armed

The gap

Four kinds of tools. None closes the loop.

Intelligence platforms

They read, you execute.

Dashboards, feeds and scores built on their curated data. The private groups where your edge actually lives aren't in any vendor's index — and when a call fires, execution is still your problem.

Bots & copy-trading

They execute, understand nothing.

Grid rules, DCA loops, or mirroring a stranger's account. Fast hands, no reading comprehension — and no answer to why a trade was taken.

Signal parsers

They only read formatted posts.

ENTRY / SL / TP templates from public VIP channels. Free-form discussion is invisible to them, and nobody is accountable for whether a source is actually any good.

AI trading agents

Black boxes trade on vibes.

Unverifiable PnL, custodial wallets and scraped sources hide what drove an action. Fenestra is the glass box: every signal carries evidence, coordinated-amplification detection checks whether confirmation is independent, every source has a verified track record, and every action receives a risk-gate verdict.

Nobody connects your sources to your execution with accountability in between. That's the layer Fenestra builds.

Pipeline

From chatter to a guarded order.

One pipeline, four stages — with you in the loop by default.

01

Connect your sources

Telegram runs through your own account session. X, Discord, Reddit, RSS and browser-extension capture plug in the rest. You choose every source and provide the required per-user credentials or authorized access — nothing comes from a vendor's index.

Telegram · liveX / Discord / Reddit / RSS / browser extension · live
02

LLMs read the discussion

Not just formatted signals. Entries, stops, targets, the thesis behind them — and follow-ups like "move SL to entry" or "close half" — are extracted into structured, auditable data with evidence spans linked back to the original message.

Extraction + lifecycle · liveNarrative & sentiment · in development
03

Every idea is validated and sized

Source track record, risk/reward geometry, signal freshness, chase distance, portfolio-level risk gates. Position size follows the source's proven edge with capped-Kelly sizing — unproven sources trade at zero.

Multi-layer validation · live
04

You approve. It executes with guards.

Default mode: the agent proposes, you tap approve. Orders go out idempotent, with target and stop-loss protection attached atomically in the same request, then verified on the venue. Full autonomy stays a deliberate, per-user opt-in.

7-venue paper/testnet fleet · liveLive money: KuCoin only
Pulse live feed

One review surface.

Pulse composes fresh signals, evidence, source scores, protective checks and approval state without creating another disconnected feed.

Deployable named agents

Each job has its own limits.

Bind an agent to chosen sources, schedule and per-agent risk knobs; shadow, paper and off modes remain explicit, and nothing auto-enables.

Read-only MCP

Verified intelligence travels.

Other agents can consume our verified intelligence, never our execution. MCP exposes only allowlisted, read-only tools.

Verification trust layer

Claims keep receipts.

Verified snapshots carry proof hashes, with optional Sui/Walrus proof anchoring. No-survivorship accounting keeps losing calls in scorecards; walk-forward checks expose selection bias when signal groups delete losers.

Safety layer

Built for real money. Verified by code, not promises.

Everything below is implemented in the live execution path today — not a roadmap slide.

Kill switch

One action halts all execution. A watchdog trips it automatically if the pipeline heartbeat goes stale.

Atomic TP + SL

Entry, target and stop-loss protection travel in a single exchange request. No naked entries — not even for a second.

Post-entry verification

Seconds after a fill, the venue is re-checked for both protective orders. Missing protection? The position is closed automatically.

Idempotent orders

Deterministic client IDs make double-submission impossible; interrupted submissions are recovered by ID, never re-sent blind.

Dry-run first

Every live order is preceded by an exchange-side test order that validates parameters before capital moves.

Hard caps in code

Max position size, open-position count, daily loss limit, leverage and margin mode are enforced before any order is built.

Arm-per-symbol

Execution is armed only for the specific market being traded, and re-guarded immediately after.

Full audit trail

Every message, extraction, decision, order and LLM call is recorded with timestamps and reasoning.

Guarded agent pattern, native. The pattern the largest retail broker now ships for agentic trading — separate limited account, order preview, manual approval and instant disconnect — is Fenestra's native architecture.

It never trades its own opinion. Fenestra executes only what you configured and approved, on your own exchange account, with API keys you can revoke at any time. And every proposal ships with its reasoning, evidence spans and a complete decision log — the audit chain regulators increasingly expect from automated trading tools is the default output, not an afterthought.

Source scoring

Every source earns its size.

Signals are cheap. Track records aren't. Fenestra attributes every closed trade back to the source that suggested it and keeps a living score per source, per market, per direction and horizon.

  • Bayesian win-rates — small samples can't overclaim; confidence builds with evidence.
  • Expectancy in R — sources are judged by realized risk-adjusted outcome, not vibes.
  • Memecoin caller verification — your memecoin callers get a track record: verified 2x-rate and rug-rate on realized outcomes, not follower counts.
  • Sizing follows proof — capped-Kelly scaling by source score; negative-edge sources are sized to zero automatically.
  • Your moat compounds — over time your private source portfolio becomes an asset nobody can copy.
illustrative
SourceHit ratenExpectancySizing
tg / momentum-desk61%47+0.42R×0.80
tg / private-alpha #255%31+0.18R×0.45
x / @macro_list48%22−0.05R×0.00
discord / degen-lounge6building…×0.00

Landscape

Where Fenestra sits.

Capability Execution terminals
(Axiom / Padre-class)
All-in-one + public intel
(Freeport-class)
Social-latency snipers Memecoin sniping terminals Signal groups / parsers Black-box AI agents Fenestra
Your own private sourcesnopublic data only: filings, X, mediapublic social flowwallet and social feedsselected channelsscraped or vendor-selectedyes — you connect them
Understands free-form discussionnopublic-feed contextlatency firstspeed and token flow firstformatted posts mainlyopaque model-dependentLLM extraction + linked evidence
Reasoning behind every actionmanual executionpublic contextspeed triggerstoken and wallet checkssignal text onlyblack boxthesis + evidence + decision log
Caller accountabilitynono caller-level outcomessocial proofwallet activityclaimed or backtested statsunverifiable PnLrealized 2x-rate + rug-rate + proof hash
Protective checksexecution controls varyterminal controlsspeed over reviewhoneypot and rug checks varybasic TP/SLagent-definedrug checks + caller score + paper-first gates
Who sets sources and riskuser tradesvendor feed / user exposureuser follows walletsvendor feed / user walletchannel owner / parser rulesagent or vendoryou — own sources, limits and approval mode

Categories describe typical products in each cluster as of mid-2026 — individual tools differ. No performance comparison implied.

Early access

The pilot cohort is deliberately small.

Closed-pilot access still opens in small waves. Once invited, use the shipped AI onboarding wizard: paste your sources + a goal, review the draft setup, then confirm. Nothing auto-enables; source connections, agent modes and risk policy remain under your control.

No spam, no sharing. You'll hear from us when your wave opens.

Process transparency

Watch machine do work.

Real pipeline stages, not performance theater. This view shows what system did — never prices, directions, source text, outcomes, or advice.

FENESTRA / PIPELINE illustrative
ingestedsocial · Social sourcewatch
extractedsocial · Social source · 3.1swatch
risk_checkedsocial · Social source · 6.2sblocked · policy