We debate. You decide.
The institutional research desk, in software. 15k names, read live, every day.
Scroll for how it works, or read today's memo now →
Institutional research memos, on every name you follow: computed, not guessed, rewritten as events move.
Under the hood, no magic: deterministic modules compute every number, and a computation cannot hallucinate. The language model, local, only writes the prose.
No advice, no execution, no verdicts. Measured states, historical reactions, named tensions. You decide.
Your data never leaves our machines: client processing is 100% in-house, local.
Read today's memo, whole and free →
Every day, the desk opens its richest dossier in full: our standing demo, and it renews itself.
Every day, our desk reads one name in full. 8 investor archetypes debate it. Today: AAPL.
You decide.
AAPL · 2026-07-10 · conviction is how strongly a framework holds its read, not a probability of outcome
This is not a mockup: these stances were computed today.
Three desks, one machine.
Parsec Editorial is our newsroom: it reads all the data as it arrives, runs its own research against each symbol's news, and weaves the links between events nobody has time to connect.
Parsec Research is the research desk: living memos, a panel of archetypes seated by the day's story on each name, and PR_Thesis: your own theses put to the debate. All of it returned as one legible memo: the essential at first glance, the depth when you want it. For B2B desks: your dedicated archetypes, shaped on your rules and your methods; a monthly re-bench keeps them current with your decisions.
Parsec Investment is the investment desk, for regulated firms only: the descriptive reading of a book, mandate by mandate, as close as possible to who you are. Exposures with evidence, allocation models labelled by name, stress as historical reaction. So you decide with everything in view. Served on dedicated confidential infrastructure.
The editorial desk feeds research; research feeds investment. Same substrate, same discipline, three doors.
We do not replace your analysts. We give you back your hours.
A memo is already written when you sit down: your magic starts where the memo ends, not where the data begins. Your quants, your PMs, your advisors work from one common base, with the debate already laid out. What is left for you is to dig where we have cleared.
The product does not replace judgment: it frees the time you need to do what you do best.
Hallucination is not forbidden here. It is impossible by construction: about 90% of the memo is computed outside the language model, and the model only narrates what has already been computed, never the reverse.
Every build is checked before it ships: numerical fidelity and provenance, verified against a reference oracle. The prose is then audited, claim by claim; what has no grounding does not pass. Today's memo carries its badge, printed on it:
Because what is not measured is said as such. On today's memo you might read: “13F ownership reads 101 days behind this memo.” The absence matters as much as the information: we print it.
Quality has a lineage, memo by memo, watched continuously. Pick a memo: we show you where every figure comes from. And the system grades itself: every past reading is scored against what actually happened, and calibration is redone as it walks. The more the desk produces, the larger its calibration base.
Compliance is not a proofread, it is the architecture: four layers, a scrubber, closed enumerations. Advice is not forbidden here either: it is impossible, even by accident.
Every piece of data here exists twice: as it was, and as it was known. Ask what the desk said on May 5th: it is a query, not a project.
Walk-forward holds by construction: the system cannot look at the future, even by mistake. There is no discipline to respect, there is no button to cheat with.
Every call carries its audit stack and its hash: what you read replays bit for bit.
And the memo is alive: it rewrites itself when information arrives, not on a schedule. A filing lands, a force flips, the memo remakes itself.
A memo that concludes on its own has already taken your decision from you. Here, the memo debates.
On the bench: more than sixty investor archetypes. Five to eight are seated per memo, and the casting is never the same: it follows the detected narrative, the asset class, the day's news.
Five lenses always sit: quantitative, macro, tail-risk, behavioral, fundamental. The guests rotate with the story: momentum, value-trap, bubble, regulation. Specialists step in by asset class. Not counting yours: the archetypes we build for B2B clients, seatable in every memo and in PR_Thesis.
Every voice arrives armed: an enumerated stance, evidence verified figure by figure against the dossier it actually read, watchpoints with numbered thresholds, and its explicit disagreement with the rest of the panel. A testable framework, not an opinion.
The debate closes without complacency: stance distribution, axis of divergence, pivot condition. Deterministic.
For B2B desks: your method, installed as an archetype, benched before delivery, debating your theses and applying your rules in every memo, every day.
And the universe is not a fixed catalogue: tens of thousands more names are onboardable on demand. Ask for a name, it joins the universe. Coverage is a consequence of ingestion: the more the desk ingests, the more it covers, mechanically.
Depth is a working desk's depth: the inverse DCF, which replaces fair value with the only honest question (what does the price imply, against what was delivered) · historical analogs with n= and ranges, reference class labelled · earnings transcripts digested: the guidance verbatim, the management's tone, the directness of the Q&A · options, 13F filer by filer quarter over quarter, insiders, congress, seasonality · the tail map, regimes, structural forces named by their mechanism · The Tape, fractal supports and resistances, the peers mesh · kill-paths, cross-lens catalysts, the glossary in context.
The filament: editorial feeds research, research feeds investment.
You entrust us with your data, your theses, your notes. Protecting them and respecting them is a duty, not an option.
Your data never touches a third-party model: no OpenAI, no Anthropic, no public cloud. Inference runs on our own machines, on our own models, locally. You are in our house, not at a vendor's.
For clients who prefer their data on their own soil: a dedicated server can be provisioned on demand, in the jurisdiction of your choice. Same architecture, same guarantees, where you want them.
The three lives of the data: in transit, ephemeral and encrypted; at rest, encrypted, keys held by the client; outside, never. External APIs are an inbound flow, nothing goes back up.
Two-database pseudonymization: whoever reads your processing data does not know who you are; on exit, the link is erased irreversibly. Passkey and magic-link: no password to steal. A confidential-computing node for regulated firms.
The compliance report reaches you monthly, pushed: you do not have to ask for an audit, it is on us to prove you are safe.
An analyst-day runs $1,200 to 3,000 and reads one name: a part of what a single memo holds. The desk keeps 15,777 names current, and rewrites about 2,300 a day, as events arrive.
Rebuilding this on API calls would cost hundreds of thousands a month, and it would hallucinate: without the deterministic substrate, the model invents. That substrate took years to build, it runs locally, and it is not for sale.
The rate card is published, no haggling. Grandfathered for life: your entry price never rises while your subscription stays active. Trying before subscribing: every one-off purchase deducts from your first month, for 30 days. Hard quotas: no surprise bill, ever. Referral: a pack of 10 PR_Thesis for each side, stackable.
Parsec is fixed-price, no bad surprises, so your budget stays where it is fully useful.
The product itself is honest: every memo says what it does not know, what it could not measure, what it is missing. “Not measured in this vintage” gets printed, not hidden. An absence is information, not a weakness.
We print our gaps, then spend every day closing them.
This desk runs in alpha, and says so. Bugs are hunted and rewarded: a real bug we had not already logged earns, your pick, a month of the PRO desk or a pack of 10 PR_Thesis. The alpha to beta to release ladder is public, with its criteria. B2B tenants vote on the roadmap for the life of the product.
The listed prices are not final: it is written on them. The track record is not told in marketing: it is demonstrated under NDA, by live replay, on the date you choose.