Platform — Every Module, Screen by Screen

DashboardThe morning surface

One screen that answers what needs attention.

What is happening today, what work is open, and what moved overnight — on one page, before you have opened anything else.

Every item on this screen names the client it belongs to and the program element it sits under, so the queue reads as a book of business rather than a list of alerts.

The Capiro dashboard: items that need attention, a day-to-week calendar, the work queue, and a filterable intelligence feed

Dashboard · All clients · Mon Aug 10

The things that need you today

Critical, deadline and hearing items, each carrying its client and its program element rather than a bare headline.

Now, today, tomorrow, the week

One column, four horizons, and a live marker on the current time. No separate calendar to reconcile.

Deadlines carry their source

Federal Register, SAM.gov, HAC‑D. The origin travels with the date, so a date can be checked rather than trusted.

Open work names its blocker

Draft ready, in progress, needs sources. The state is the next action, so nothing sits in a queue unexplained.

MeriAI chief of staff

Work starts in one place.

One start-work surface holds the skills and the prompts. Both write an editable intent into the composer, and the composer is the only place work begins.

Meri reads Capiro intelligence, your workspace records, institutional memory and connected applications — so you do not re-explain your clients on every question.

Meri's undocked workspace: skills and prompts above the composer, with conversation context and connected sources beside it

Meri · New thread · Portfolio scope

Skills and prompts, one surface

Two tabs over the same surface. Both fill the composer; neither auto-sends.

A skill that needs a client says so

At portfolio scope it opens the scope selector first, then fills the intent.

Three task modes, one composer

Standard, research and write share the same box, the same scope control and the same send state.

Sources state their condition

Connected, reauth needed, available — each said in words, not colour alone.

MeriSkill result

Analysis the client can audit.

A federal-market read on an outside organisation, with the chart generated from a server-side spec and every computed number marked as computed.

A research result: where the money comes from, where it touches the portfolio, and a stacked-bar chart marked as derived

Meri · Research a company · Derived figures marked

Derived is marked derived

A computed figure carries inference, not quoted, so it is never read as a source figure.

The chart renders a spec

Chart type, series and values come from the analysis artifact. The client does not choose them.

Provenance under every chart

Source datasets, row count and generated-at. Without the marker, the chart does not render.

Deep research is offered, not started

Clicking opens clarifying questions rather than launching a blind run on your budget.

PortfolioClient record

The account and its public record, in one place.

The client profile, its federal footprint, and the advocacy history behind it — each figure traceable to a filing rather than to a spreadsheet someone kept.

Lobbying disclosures, federal awards, political giving and the districts a client sits in are read from the public record as-is, then attributed on the page.

Portfolio · Client record · Overview of six tabs

Six tabs, one record

Overview, capabilities, relationships, work, intelligence and files — not six systems that disagree.

Filings carry their form and quarter

The latest filing is named down to the form and the period, so it is checkable against the source.

Advocacy history is a pipeline

Requested, submitted, sponsor, amendment, made in order, House passed — with dates, and an evidence trail behind each step.

Profile gaps are counted

One gap, named, with a sync action beside it. Meri drafts the missing fields for your review rather than filling them silently.

StrategiesWorking timeline

The plan sits on the real calendar.

Every strategy is a plan aligned to the appropriations, authorization, regulatory and procurement calendars — not a generic project board with dates typed into it.

Three fiscal years, centred on today, with the legislative, PPBE and firm-phase calendars as reference layers under the steps.

A strategy on the working timeline: three fiscal years, reference layers, typed steps, and plan health

Strategies · Working timeline · FY27–FY29

Reference layers, not plan content

Legislative, PPBE and firm phases are the context the plan is aligned to, and a partner can strip them back to steps alone.

Steps are typed

Action, gate, milestone, decision and risk each carry different fields, because they are not the same kind of thing.

Plan health is arithmetic

Ten of twelve steps have owners; two proposals need review. The gap is stated rather than assumed.

Meri reads the plan, never writes it

Ask the assistant scoped to this plan and changes come back as reviewable proposals on their own layer.

WorkspaceDocument editor

The draft and its reasoning, side by side.

The agent panel on the left, the document in the middle, and the plan that produced it on the right — so a reviewer can see the paper and the argument for it at once.

Draft, refine, review and deliver are the states of one document, not four folders. While the agent is drafting, the file cannot be edited underneath it.

The document editor: agent revisions on the left, the client paper in the middle, and the approved direction, assumptions and evidence on the right

Workspace · Document editor · Outline v2 · Six sources

Agent edits are logged and applied

Every revision says what changed and marks itself applied, so the document has a history rather than a mystery.

Every plan field says who set it

User, suggested, retrieved. The document knows which parts are yours.

Assumptions are visible and confirmable

Each one is stated in words, with confirm and reject beside it, before it hardens into a claim.

Sections declare their evidence

Word ranges and evidence counts per section — including an explicit “no evidence”.

EngagementMeetings

Every meeting arrives prepared.

Calendar-synced meetings with a drafted prep pack, Meri attendance, and a human approval before anything is used in the room.

Meetings are synced from the firm mailbox rather than re-entered by hand, so the calendar you already keep is the calendar the prep is built from.

The meetings surface: a day view with prep state on each row, beside a prep pack with ordered talking points and explicit watch-outs

Engagement · Meetings · Prep drafted, then approved

Prep state sits on the row

Prep approved, Meri attending, generate prep — the row says what is missing without opening it.

Talking points are ordered

Specific, numbered, and drawn from client context and recent activity rather than from a template.

Watch-outs are explicit

Where two records conflict, the pack says so and names the conflict before you walk in.

Approval never implies sharing

Regenerate, edit, export and approve are four distinct controls, and approving is not sending.

EngagementOutreach

Governed outreach, stage by stage.

Describe the outreach in your own words. Goal, targets, delivery, context, drafts and send are six gated stages, and one stage is in progress at a time.

An outreach campaign: the goal in the operator's own words, Meri's read-back, and six gated stages down the right

Engagement · Outreach campaign · Six governed stages

It starts from your words

Client, sponsor and audience stay optional until you choose them.

Four intents, not a form

Share an update, make an ask, invite people, or something else.

Meri reads the goal back

Topic, outcome, ask and sponsor, restated for you to correct before anything is drafted.

Nothing runs ahead of the operator

The remaining stages wait. The campaign brief is the record, on one line, editable and audited.

LegiWatchHearings

The week ahead, counted against the book.

Public proceedings on one calendar, counted against the portfolio rather than merely listed — with unattended recording, a live room while a hearing runs, and matched client moments.

Capture starts before the gavel, stops at adjournment, and delivers the transcript, the matched moments and the client impact into the session record.

The hearings calendar for one week with client matches per proceeding, a live session beside it, and scheduled unattended recordings

LegiWatch · Hearings · Aug 10–16

Counted, not just listed

The week is measured against your portfolio, and each card carries its committee and its client matches.

Postponed stays on the calendar

Struck through rather than removed, so the change itself stays visible.

Four proceeding states

Live now, hearing, markup, recording queued — each with a dot and a word.

Session history states its outcome

Recap ready, reviewed, or failed-recoverable with the cause named.

LegiWatchMorning digest

A calm summary, and what cannot wait.

What changed, which clients are affected, and which decisions are due — assembled at 7:00 AM with its source confidence stated on the page.

The Policy Morning Brief: the decision count, what changed since yesterday with provenance, decisions due with owners, and brief controls

LegiWatch · Morning digest · Source confidence stated

It opens with the decision count

Four decisions today, and the highest-impact issue named first.

Each change carries its provenance

Client, severity and procedural path, in machine-readable form under every line.

Unknowns are retained, not dropped

Explicit unknowns are kept in the brief rather than smoothed over.

Generating a brief is not sending it

Export, send and schedule are three distinct actions.

Program ElementsDefense budget intelligence

Defense funding, line by line.

Budget lines followed from request through Congress to enacted outcomes, each with its evidence state and a recommended action — server-paginated, so a filter changes the answer, not the page.

A position carries its stage: a number without a stage means nothing. Deltas are shown in dollars and percent, because the two answer different questions.

Defense budget lines with PE code, request, current position and stage, delta, evidence state and a recommended action per line

Program Elements · Explore · FY27

Every line is its PE code

Program, code, component and appropriation type on one row.

Evidence state is separate from the number

Verified, conflicting, partial, review — never folded into the figure itself.

A recommended action per line

Protect funding, restore funding, monitor, review mark — attached to the line, not to a memo.

Multi-year trend in place

Five fiscal years of the same line without leaving the row.

Program ElementsMovement

What changed, and what it means.

Official budget movement with materiality, the stage that produced it, and a written decision posture — so a partner reads a position, not a diff.

Budget movement: a movement tape of official position events, a distribution of affected clients, and a written decision posture

Program Elements · Movement · FY27

“Official change” is a label

Official change, new stage and no material change are three different events, not three words for one.

Silence is a finding

No material change is reported, rather than left out as an absence of data.

Distribution, not just a list

Bubble size is affected clients; position is movement against request.

The next window is stated

The advocacy clock is explicit, so the posture comes with a date attached.

District NexusCongressional footprint

A footprint you can defend.

Verified client facilities, public funding and representation translated into a congressional footprint — with every gap in the data named on the page.

Real district geometry, with CONUS, Alaska, Hawaii and Puerto Rico as separate views fitted to the client footprint rather than squeezed into one frame.

Client footprint by congressional district: districts, employees, contracts and grants, a real-geometry map, and a scored nexus table

District Nexus · 119th Congress

Nexus strength is a scored band

Zero through 80+, and “no verified nexus” is kept as its own state rather than reported as a zero.

Blank is not zero

Where districts are not fully reported, the figure says so instead of guessing.

Unplaced funding is called out

Contracts and grants that could not be tied to a district are named, not quietly dropped from the total.

Coverage sits beside the score

Every district carries its nexus score and how much of it is actually evidenced.

DirectoryCommittee staff

Committee structure, from the source.

Authorizing and appropriations staff as a source-backed office structure, with the lines claiming only what the record supports.

Personal-office staff, committee staff, Department of War and state legislators sit behind one canonical identity per person — the same person across engagement, outreach and the quick log.

Directory · Committee staff · Hierarchy or roster

Bipartisan leadership is a relationship

Chair and ranking member sit at the same level, joined by a dashed line.

The legend defines the lines

Solid is formal structure, dashed blue is a working relationship, muted is unfilled.

It claims only what the source supports

Office placement, never an implied person-to-person reporting line.

Unfilled seats stay visible

A pending assignment is shown muted rather than omitted from the chart.

How it is built

Nothing on the screen claims more than it can show.

The same rules run through every module above. They are the reason a partner can put a Capiro screen in front of a client.

Provenance

Every figure names its source.

Dates carry the register they came from; filings carry their form and quarter; charts carry their datasets, row count and generated-at.

Derivation

Computed is never quoted.

A derived number is labelled as inference. If the server returns no reading, the strip is omitted rather than filled in.

Gaps

Missing is a state, not a zero.

Blank stays blank, unknowns are retained, unplaced funding is named, and “no material change” is reported as a finding.

Governance

The agent proposes; a person applies.

Meri drafts, reads back and proposes. Approval, export and send are separate actions, and every stage is gated and logged.

Integrations

Works with the tools you already use.

Capiro connects to your mailbox, calendar, CRM, meeting notes and files, so it has full context the moment you turn it on. No migration, and no starting from a blank slate.

Email & calendar

OlOutlookMailbox and calendar sync

GmGmailMailbox and calendar sync

CRM

SFSalesforceContacts and accounts

HSHubSpotContacts and pipeline

Files & storage

SPSharePointDocuments and files

ODOneDriveDocuments and files

GDGoogle DriveDocuments and files

Meeting intelligence

OtOtter.aiMeeting transcripts

FiFireflies.aiMeeting transcripts

KrKrispMeeting transcripts

Communication

SlSlackAlerts and notifications

MTMicrosoft TeamsAlerts and notifications

Project management

AsAsanaTasks and deadlines

MoMonday.comBoards and workflows

Under the hood

An enterprise platform purpose-built for the legislative environment.

Capiro runs its own data pipelines, agentic workflows and relationship graph — the parts general enterprise tools are not designed to support.

Intelligence engine

46 purpose-built AI agents.

One orchestrating agent managing 46 sub-agents, each with a defined scope, alongside 17 ML models running continuously — gradient-boosted classifiers, NLP attribution, time-series forecasting and graph centrality analysis.

Federal data

75+ live federal sources. 50M+ records.

Congress, DoD, FEC, SAM.gov, USASpending, the Federal Register and LegiStorm, continuously ingested, with 1,536-dimension semantic embeddings so a program, precedent, staffer or submission is findable in seconds.

Relationship graph

Influence mapped, not inferred.

Committee membership, coalition paths and access routes across the legislative ecosystem, with every relationship connected to every submission across your full practice history.

Security

Tenant-isolated by default.

Multi-tenant isolated environments with no cross-organisation data access, role-based controls and a full audit log on every action. Your data never trains a shared model. SOC 2 and FedRAMP are on the roadmap.

Screen by screen.