A Work Agent for Researchers and Insight Teams

Assign Floatboat a defined research task. It can organize selected sources, build evidence matrices, compare findings, and maintain a dated research brief while researchers retain verification and methodological judgment.

Floatboat Agent

researchers and insight teams · Market transition · Evidence assignment

Working

Turn a defined question and source set into a searchable inventory and thematic synthesis.

▱ Research question▱ Approved sources▱ Coding framework

I’ll run the “Prepare a literature review” workflow, starting with the materials you provided.

Activity0/4

·Set scopeRunning…
·Collect sources
·Code findings
·Draft synthesis

Organizing context and preparing editable deliverables

＋

Ask a follow-up, add files, or ask the agent to revise the deliverable…

↑

Streaming interface preview · Click the composer to download Floatboat. No real data is processed.

Common tasks and representative workflows

What Researchers Can Delegate

Research quality erodes when collection outpaces classification, notes lose their sources, and conflicting findings are flattened into a convenient summary.

Common tasksPrepare a literature reviewBuild an evidence matrixRun recurring topic monitoringSource mapEvidence matrixSynthesis outline

Prepare a Literature Review

The problem

Source collection outgrows bookmarks

Search results, papers, reports, datasets, and archived pages quickly exceed what browser tabs and bookmarks can explain.

What the agent handles

Turn a defined question and source set into a searchable inventory and thematic synthesis.

Set scope · Collect sources · Code findings · Draft synthesis

Deliverable

A source inventory, themes, gaps, and review outline.

Human boundary: Researchers verify inclusion criteria, citations, methods, and interpretation.

Build an Evidence Matrix

The problem

Notes lose their evidence trail

A useful observation becomes unreliable when its quotation, page, method, publication date, or retrieval context disappears.

What the agent handles

Compare claims, samples, methods, dates, and limitations across a selected body of work.

Define fields · Extract evidence · Compare studies · Flag conflicts

Deliverable

A traceable matrix of findings and limitations.

Human boundary: The researcher resolves ambiguity and decides evidentiary weight.

Run Recurring Topic Monitoring

The problem

Monitoring work never really ends

New publications keep arriving, but few deserve to reopen the evidence matrix or change the current conclusion.

What the agent handles

Track selected publications and update a living brief when meaningful evidence appears.

Define watchlist · Collect updates · Classify changes · Publish brief

Deliverable

A dated update with new evidence and unresolved questions.

Human boundary: Humans approve sources and decide whether an update changes the conclusion.

A capability stack for the work

Floatboat Capabilities for Researchers

The Work Agent uses these capabilities inside a persistent workspace to read context, execute steps, and deliver results. People retain consequential decisions.

Agent Workspace

an Agent Workspace that preserves researchers project context can keep the research question, selected sources, coding framework, evidence matrix, review notes, and synthesis connected in one traceable project.

The same context stays available across follow-ups, revisions, and recurring work, so the agent can continue from an inspectable project instead of starting over in a new chat.

FloatboatMarket transition · Evidence assignment

Project files

▱ Research question

▱ Approved sources

▱ Coding framework

Agent run

Source mapEvidence matrixSynthesis outline

Multiple AI Models

multiple AI models inside one researchers project can match models to long-context reading, comparison, structured extraction, or critical review while preserving the same source set, instructions, and evidence trail.

The model can change with the stage of work while approved files, instructions, prior decisions, and deliverables remain in the same workspace. Users do not need separate provider API keys for officially available models.

FloatboatMarket transition · Evidence assignment

Auto

Automatic routing

Same workspace context

Claude Opus 4.6

GPT-6 Astra

Gemini 3.1 Pro

AI File Organizer

Normalize papers, reports, datasets, and notes locally, preserving recognizable sources and versions so the evidence trail remains navigable.

The organizer handles file preparation as a separate, visible step; people still decide which folders are in scope and verify the proposed structure before relying on it.

FloatboatMarket transition · Evidence assignment

Before

□ 1_Research question_final

□ 2_Approved sources_final

□ 3_Coding framework_final

Suggested structure

└ Research question

└ Approved sources

└ Coding framework

Skills Marketplace

Add reusable research procedures for source collection, evidence extraction, monitoring, and document production while retaining project-specific review steps.

A Skill packages a repeatable method and its expected output, giving the team a starting point that can still be inspected, changed, and governed for each project.

FloatboatMarket transition · Evidence assignment

Skill 01

Prepare a literature review

Reusable

Skill 02

Build an evidence matrix

Reusable

Skill 03

Run recurring topic monitoring

Reusable

Explore More Work Agents

Compare how a Work Agent handles real assignments across different professions.

Questions before you start

Work Agent Questions for Researchers and Insight Teams

Researchers need provenance, explicit inclusion rules, and a clear separation between source extraction and interpretation.

Is Floatboat a reference manager or academic database?+

No. Keep your authoritative databases and citation manager. Floatboat helps turn a selected source set into searchable notes, comparison tables, research briefs, and repeatable monitoring workflows.

Can it decide which evidence is credible?+

It can expose methods, dates, samples, claims, and limitations, but the researcher defines inclusion criteria, resolves ambiguity, weighs evidence, and owns the interpretation.

Can we define our own review protocol and evidence fields?+

Yes. Configure the question, databases or source set, date range, inclusion rules, extraction schema, quality checks, and output format for each project.

How do we prevent fabricated citations or unsupported synthesis?+

Require every extracted claim to retain its source location, keep unavailable evidence marked as unavailable, and verify citations, quotations, and numerical findings before using the synthesis.

Can unpublished or restricted research material be used?+

Only when permissions, agreements, and organizational policy allow it. Separate restricted projects, limit access, and review how each selected model and connector processes the material.

Will it update a literature review automatically?+

It can monitor approved sources and draft a dated change brief, but a researcher should decide which new items meet the protocol and whether they alter the review.

What is a good first research workflow?+

Choose a small, known corpus and define the extraction fields in advance. Build an evidence matrix, audit it against the originals, and only then expand the source set.

Build the source matrix before drawing conclusions

Define the corpus and extraction fields first. Keep each finding linked to its source when the synthesis moves to review.

Download Floatboat