Analyst copilots
Agents that read filings, transcripts, and expert calls, then surface what changed — with every claim linked to source.
Private AI research agents for hedge funds
Know what changed, and why it matters to your book.
Deskside connects the research you already pay for, including broker notes, alt data, expert calls, and filings, to your own notes, models, and positions. Agents watch all of it, compare every update with what you already know, and tell your team what changed. Everything runs inside your environment.
- Runs in your environment
- Never used to train models
- Never shared across clients
- You own everything we build
- ▲ Revised upCompany A · tracking above consensus
- ↺ Tone shiftMgmt language on capex softened
- ≠ Sources disagreeCard panel vs. expert call
- ⚠ CompliancePossible MNPI held back
- ✓ Brief ready3 items need your eyes
Your data, your IP, your environment.
Everything we build runs inside your walls and stays yours. Privacy and security are written into every contract.
- Runs in your environmentDeployed in your cloud or on-prem. Your data never lands on our systems.
- Never used to train modelsNot ours, not anyone's. AI providers are used only on no-training, zero-retention terms.
- Never shared across clientsStrict information barriers. Nothing from your fund is ever used for another.
- You own what we buildCode, prompts, outputs, and data are all yours, with full handoff and no lock-in.
- Deleted when we're doneReturned and deleted at the end of the engagement, with written certification.
Built by a team with decades in
Hedge funds Quantitative research Production engineeringWhy Deskside
Generic AI doesn't know what your PM needs by 7am.
Chatbots answer the questions you think to ask. Deskside agents do the reading you don't have time for, check it against your model and your data, and only bring you what changes the picture.
Every workflow is designed by a team with decades of buy-side experience. We know what an analyst does before the open, why a quant won't trust a signal without a backtest, and what compliance will ask about first.
Typical AI toolsDeskside
- A chatbot you have to promptAgents that work before you ask
- Generic answersFramed to your book and coverage
- Confident guessesEvery claim sourced, unverified items flagged
- Your data sent to a vendorRuns inside your environment
- Months of integrationLive on your coverage in weeks
What we build
From the research desk to the back office.
We focus on the work that eats your team's hours and where a well-built tool compounds every day.
Alt-data pipelines
Turn messy vendor feeds, web data, and unstructured documents into clean, queryable signals your quants can trust.
PM decision support
Morning briefs, position monitoring, and natural-language access to your own risk and P&L systems.
Ops automation
Reconciliations, trade breaks, investor reporting, and DDQ responses — handled by agents, reviewed by people.
Compliance tooling
Communications review, restricted-list checks, and audit trails built so your CCO is comfortable signing off.
Internal AI platform
Secure model access, evaluation, and guardrails so your own team can build the next ten tools themselves.
Automatic report summaries
Every report read, compared, and summarized for you.
Broker notes, alt-data updates, and expert-call transcripts arrive faster than anyone can read them. Our agents read each one the moment it lands and tell you what changed: which estimates moved, how the tone shifted, and whether it matters for your position.
New report · Company A · Q3 tracking update
Tracking moved from in line to above consensus
Estimate revisions vs. the prior report
| Metric | Prior | Now | Consensus | Change |
|---|---|---|---|---|
| Revenue growth Y/Y | +6.1% | +7.4% | +6.3% | ▲ 130bp |
| Active customers | +2.0% | +2.2% | +2.1% | ▲ 20bp |
| Spend per customer | +4.0% | +3.6% | +4.2% | ▼ 40bp |
Tone inflection
- “demand remains steady”→“demand is accelerating into the quarter”
- “no change to our outlook”→“upside to our prior estimate”
CautiousNeutralPositive
What it means for the book
Supports the long. Growth is running ahead of the Street, but spend per customer slipped for the second report in a row. That's the leg to watch into the print.
Revisions, not just readouts
Every new report is lined up against the last one from the same source and against consensus, so you see which way estimates moved, not just where they are.
Tone and language shifts
Changes in wording like “steady” to “accelerating,” or “in line” to “tracking above” are pulled out side by side. Softening language gets flagged even when the numbers haven't moved yet.
Across sources, not one at a time
When a broker note, an alt-data update, and an expert call disagree about the same company, the summary says so and shows where each one stands.
Framed to your positions
Each summary ends with what it means for the book: confirms, cuts against, or doesn't matter. It is delivered to email, chat, or your research notes.
Live call intelligence
Flags while the call is still happening.
We connect to the note-taker your team already uses, such as Granola, and read the transcript as it's written. During expert calls, earnings calls, and management meetings, the agent checks what's said against your model, your data, and what was said before, and flags what matters in real time.
-
Expert
“Pricing held through the spring, but we started seeing a lot more discounting in June.”
Contradicts earlier commentary. A prior call in March said “no discounting this year.” -
Expert
“I'd say fifteen, maybe twenty percent of our accounts were testing the new entrant.”
Differs from the data. Panel share for the entrant is about 11%. Worth asking how accounts are counted. -
Analyst
“Is that mostly enterprise, or smaller accounts?”
-
Expert
“I probably shouldn't say this, but the numbers that went to the board last week…”
Compliance: possible MNPI. Compliance alerted; this segment is held out of the notes. -
Suggested follow-up. “Did the June discounting carry into July, or was it end-of-quarter?”
Checked against what you already know
Every claim is compared in the moment with your model, your alt data, and prior calls with the same company or expert. Contradictions and new numbers are flagged, not buried in a transcript.
Compliance built in
Language that suggests material non-public information triggers an immediate alert and keeps that segment out of the shared notes, following your firm's policy.
Better questions, in time to ask them
The agent suggests follow-ups while the expert is still on the line, based on the gaps between what they said and what your data shows.
Notes filed before you hang up
When the call ends, you get main takes, model-relevant numbers, inconsistencies, and verbatim quotes linked to the transcript, filed to the right company in your research notes.
Example builds
Tools we've run on a live book.
These are examples of systems we've built and used ourselves, day after day. Every client version is built from scratch around your coverage, your data, and your process.
- Macro & tapewhat the market is pricing today
- Company newsonly what changed since yesterday's brief
- Broker actionsrumors checked against the source
- Channel checksnew research across your portals
- Insider filingscoded by signal strength
- Technical levelsnames sitting on key moving averages
- Data readsalt-data vs. consensus
- Three pitchesbull and bear cases, checked for consistency
- Catalyst calendarconfirmed vs. estimated dates
- Data-integrity noteswhat was stale, missing, or unverified
The pre-open brief
Runs every morning before the open. It sweeps overnight news, broker actions, insider filings, research portals, alt-data releases, and technical levels across the whole coverage list. Then it frames every item against the actual book: does this confirm the position, or cut against it?
The part PMs notice is what it won't do. When a broker upgrade shows up in a feed but can't be confirmed at the source, the brief says so and leaves it out. Stale data is labelled stale. Nothing gets a number the source didn't give.
Call notes in minutes
Expert calls, conferences, and podcasts are transcribed privately and turned into main takes, model-relevant numbers, and contradictions with what management said before. Every line is footnoted to the transcript, and garbled figures are flagged instead of guessed.
Expert-series synthesis
Reads across a whole series of expert or agency panels at once. It accounts for each panel's built-in bias and focuses on what's tradeable: which way estimates are moving, and how names rank against each other.
Alt-data vs. consensus
Archives every vendor release, lines up each KPI against consensus, and tracks whether a beat or miss is persisting or fading. It tells a fresh read from a re-published old one, so nobody trades on last month's data.
Earnings-call diffs
Each transcript is compared line by line with last quarter's: guidance language, new risk wording, questions management dodged, and shifts in tone, ready minutes after the call ends.
Research that files itself
Filings, broker research, models, alt-data, and call notes land in the right folder per name, automatically. Thesis patterns are tracked over time, including the ones deliberately rejected, and why.
Insider and event scans
Checks regulatory filings across the coverage list, separating scheduled selling from real open-market buys, and ties each one back to the position it affects.
Alt-data agents
Every dataset you license, working every day.
Funds pay for card panels, receipt data, app and web traffic, and channel checks. Then analysts have time to open a fraction of it. We build agents that take in every release, check vendors against each other, and turn raw panels into the analysis you'd otherwise wait on a data team for.
Signal triangulation
Compare what several vendors say about the same KPI, adjust for each panel's known bias, and flag when they disagree ahead of the print.
Automated cohort analysis
Rebuild retention, repeat-rate, and spend-per-customer curves by acquisition cohort every time new panel data lands.
Nowcasts vs. consensus
Turn panel reads into KPI estimates with backtested error bands, and track the gap to the Street into earnings.
Panel health checks
Catch shrinking panels, merchant drop-outs, methodology changes, and rebases before they show up as a false signal.
Dataset trials
Backtest a new vendor against reported numbers before you sign the contract, not after.
Point-in-time by default
Every series is stored as it was known on the day, so backtests don't quietly peek at the future.
Retention by acquisition cohort, rebuilt with every data drop.
Months since first purchase → ← Acquisition cohort| Source | Read | Chart |
|---|---|---|
| Card panel | +4.1% | |
| Receipt panel | +3.4% | |
| App activity | +5.0% | |
| Bias-adjusted blend | +4.0% | |
| Consensus | +2.8% |
Sources agree: tracking ~120bp above consensus. Panel health normal.
Analyses on autopilot
The work a data team would do, rebuilt with every release.
When new panel data lands, the agents rerun the analyses your team relies on and flag what moved. Three examples, shown with synthetic data:
Newer cohorts are holding on longer
Share of each acquisition cohort still active, by months since first purchase.
- Q1 '25
- Q2 '25
- Q3 '25
- Q4 '25
- Q1 '26
What the agent flags: month-3 retention is up about 4 points across the last four cohorts, so growth is coming from customers who stay, not just new sign-ups.
View data table
Who is the new entrant taking share from?
Share of category spend by month, from a transaction panel.
- Incumbent A
- Incumbent B
- New entrant
- Other
What the agent flags: the entrant reached about 11% share within a year, taking roughly 1.6 times as much from Incumbent B as from Incumbent A.
View data table
Three ways to project the quarter, blended and checked against the Street
Revenue growth year over year. For the quarter in progress, each extrapolation method is shown next to the blended estimate, consensus, and the guidance range.
- Reported
- Our estimate
- Consensus
- Extrapolation methods
- Guidance range
View data table
All charts use synthetic data for illustration. No client, vendor, or company data is shown.
Proactive agents
Agents that work before you ask.
Most AI tools wait for a prompt. We also build agents that run on their own: they watch your book, your data, and your inbox around the clock, and come to you when something needs attention.
- Event-driven. A filing hits, a price gaps, a vendor file arrives late, and the agent is already working.
- Scheduled. Briefs, reconciliations, and reports ready before the team sits down.
- Where you already work. Alerts go to Slack, Teams, email, or Symphony, with sources attached.
- On your rules. You decide what they watch, when they escalate, and what always needs a human sign-off.
- Filings
8-K filed on a top-10 position: CFO departure. Summary and prior-CEO-change precedents sent to the PM.
- Data
Card-spend panel arrived with 18% fewer merchants than usual. Signal refresh paused, quant team notified.
- Brief
Pre-open brief ready. A feed-reported upgrade couldn't be confirmed with the broker, so it was flagged and left out.
- Ops
2 trade breaks with the prime broker matched to a booking error. Fix drafted, awaiting ops approval.
- Risk
Sector exposure crossed the limit you set. PM and risk alerted with the positions driving it.
How we deploy
Live on your desk in weeks, not quarters.
-
Week 0
Sit with the desk
We shadow your analysts, PMs, or ops team for a few days. We map where time goes and pick one workflow where AI can pay for itself quickly.
-
Weeks 1–2
Ship a first version
A working tool in the hands of real users — inside your environment, on your data, with your security team in the loop from day one.
-
Weeks 3–8
Iterate at the desk
Daily feedback, tight loops, and evaluations against your own examples. We harden it until the team relies on it without thinking.
-
Ongoing
Hand off or expand
You get the code, documentation, and a team that knows how to run it. Many clients then point us at the next workflow.
Who we work with
Hedge funds first. Anyone with a desk, second.
Core focus
Hedge funds
Fundamental long/short, multi-manager pods, macro, and quant shops. We understand the pace, the secrecy, and why a tool that's right 95% of the time isn't good enough.
Asset managers
Research scale, client reporting, and RFP/DDQ automation.
Family offices
Lean teams that need leverage across a broad portfolio.
PE & beyond
Diligence, portfolio monitoring, and any team drowning in documents.
Our commitments
Built for firms where trust is the product.
Your data stays yours
We deploy inside your cloud or on-prem. No training on your data, no shared tenancy, no surprises for your CISO.
Every answer has a source
Outputs cite the filing, the page, the cell. If the model can't back it up, the tool says so.
Compliance in the room
We work with your CCO and security team from the first week, not the last. Logging and audit trails are built in.
You own everything
Code, prompts, evaluations, and infrastructure are yours. No platform lock-in, no per-seat tax.
Get started
Tell us the workflow that eats your team's week.
You'll talk to the engineers who'd do the work, not a salesperson. We'll tell you honestly whether AI can help, and where we'd start.