Agent Studio/ closed-loop factoryWhat is Studio?

A factory for capital-grade agents, continuously re-tuned against the live tape.

The Studio formalizes our move from a static toolset to a modular Agent-as-a-Service architecture. Each agent is shipped with a single mandate, then re-trained on market feedback and desk telemetry — every release closes the loop.

14
Agents in production
2,184/24h
Daily handoffs
23.2sp50
Closed-loop latency
94.6%90d
Desk retention
01Production workflow

A four-phase line, instrumented end-to-end.

Discovery feeds Development; Development hardens into Orchestration; Orchestration emits Output. Then telemetry from desks rewrites the priors — the line never stops.

01 · Discovery

Ingest the live tape

Continuous RSS, REST, and on-prem feed adapters draw market data, sovereign filings, and desk notes into the staging lake.

Core technology
RSS / API ingestion
BloombergReutersIMFBISInternal desks
02 · Development

Train the priors

Bayesian priors are calibrated on 10y of regime data; the Neural Matcher learns expert-to-peer alignment under sparse labels.

Core technology
Bayesian priors · Neural Matcher
StanPyMCPineconeEmbeddings
03 · Orchestration

Wire the handoff

Each agent is a node in a typed DAG. LangGraph supervises chain-of-thought; n8n carries IO contracts to humans-in-the-loop.

Core technology
LangGraph / n8n handoff
LangGraphn8nZod schemasWebhooks
04 · Output

Publish the brief

CapitalIssues-branded SWOT and SIU templates render the final artifact — narrative, three data tables, no bullet points.

Core technology
CapitalIssues SWOT / SIU
MarkdownPDFNotionSlack
Closed-loop topology
feedback · retune
Market feedsRSS · API · desksBayesian corepriors · regimeMacro-Scoutagent · 01Neural Matcheragent · 02Editorial Synth.agent · 03SIU briefoutputfeedback · retune
02Agent catalog

Three specialists. One mandate each.

Every agent is built for one job, with one measurable output. Generalist behavior is delegated to the orchestration layer, not the unit.

AGT-001

Macro-Scout

Pulls sovereign-risk telemetry every five minutes and scores early-warning signals against a calibrated baseline.

Key feature

Early-warning sovereign-risk scoring on a 0–100 stress index.

Triggers
  • CDS basis > 80bp
  • Reserve coverage < 3mo
  • Energy Δ > 8% / 24h
Output contract

Stress index + 5-line situational note

AGT-002

Neural Matcher

Aligns the requesting analyst with the closest expert peer in our internal corpus, weighted by topic, regime, and prior accuracy.

Key feature

Expert-to-peer alignment with a confidence-weighted ranking.

Triggers
  • New ticket opened
  • Confidence < 0.55 from prior agent
Output contract

Ranked expert list + intro brief

AGT-003

Editorial Synthesizer

Drafts a 1,000-word Barron’s-style narrative report with three mandatory data tables and no bulleted lists.

Key feature

CapitalIssues-branded prose with strict table contracts.

Triggers
  • Scout + Matcher chain complete
  • Manual desk request
Output contract

SIU brief — 1,000w narrative + 3 tables

0324-hour market shift demo

From prompt to publishable brief in under 25 seconds.

user input“Closure of the Strait of Hormuz”
  • 01
    Scout

    Pulls 12y of Hormuz closure precedent + current tanker telemetry, OPEC quotas, Iranian fiscal data.

    4.1s
  • 02
    Analyst

    Runs a Bayesian simulation across 10,000 paths over a 30-day horizon — prices, CDS, fiscal balances.

    11.8s
  • 03
    Writer

    Drafts the 1,000-word narrative, populates three regional tables, hands off to desk for review.

    7.3s
Total closed-loop23.2s
SIU brief · preview
24-hr · regional impact

Hormuz closure — fiscal & energy sensitivity, Δ vs. 30d baseline

RegionFiscal ΔCDS bpEnergy Δ
GCC / Saudi+1.8%+22+11.4%
Iran−4.6%+184+9.2%
Oman / UAE+0.3%+41+8.7%

“A full closure widens GCC sovereign CDS by an average of 22bp and lifts spot Brent sensitivity into double digits. Oman and the UAE absorb the smallest fiscal hit — their export routing is the most diversified and the…”

Editorial Synthesizer · 1,000w · 3 tables
04Token-plus-seat pricing

Pay for the seat. Pay for the tokens. Nothing in between.

Professional
$250/ mo

Single analyst, three agents, one desk seat.

  • 1 desk seat + 250k tokens / mo
  • Macro-Scout + Neural Matcher
  • Editorial Synthesizer (read-only)
  • Custom agent provisioning
  • Dedicated re-tune cadence
  • On-prem deployment
Request access
most desks pick this
Enterprise
$1,200/ mo

Most desks pick this — full catalog, shared budget.

  • 10 desk seats + 2M tokens / mo
  • Full agent catalog + handoffs
  • Editorial Synthesizer (read/write)
  • Custom agent provisioning
  • Dedicated re-tune cadence
  • On-prem deployment
Request access
Strategic / SIU
Custom

Bespoke agents, sovereign-grade SLAs, on-prem option.

  • Unlimited seats + token pool
  • Full agent catalog + handoffs
  • White-labeled SIU briefs
  • Custom agent provisioning
  • Dedicated re-tune cadence
  • On-prem deployment
Request quote
Provision the Studio

Get an access link, a sandbox tenant, and a 30-minute walkthrough.

We provision a sandbox tenant within one business day. The walkthrough covers agent provisioning, output contracts, and how the closed loop is observed in production.

We reply within one business day. No marketing list.

SAgent Studio / closed-loop factory

A surface of the broader CapitalIssues platform. Agents are shipped, observed, re-tuned, and shipped again.

Surfaces

Studio

Status

Studio · v1.0closed-loop · always retuning

GDPR · Privacy & cookies

Your data, your rules.

We use strictly necessary cookies to run this site and, with your consent, anonymous analytics to improve it. You can change your mind at any time. Read more in our Privacy notice and Cookie policy.