An AI research assistant with sources.
Ask a question worth investigating. Ghost builds a focused plan, gathers relevant sources, and turns the findings into a report you can inspect, save, and use in the rest of your work.
Multi-source research · Evidence and gaps · Private beta for Mac
Should Atlas change its onboarding?
- Question
- Which parts of onboarding need a closer look?
- Plan
- Compare product documentation, recent customer questions, and the decision from the last review.
- Output
- A source-linked brief, open questions, and a proposed next step.
From an open question to a sourced brief.
Follow the Atlas team’s onboarding question through a focused investigation, a source review, and a brief that can become the next piece of work.
Give the question a useful boundary.
Start with the decision, audience, and what evidence would change your mind. ghost can turn a broad topic into a focused research brief.
What should we change in onboarding?
- Decision
- Which friction should the team investigate first?
- Scope
- Current product documentation and customer feedback
- Output
- Sourced brief with open questions
Choose sources for the question.
Public research and your saved work are different evidence paths. Ask ghost to use the relevant material, with source access and missing coverage stated clearly.
- Public material
- Original documentation and relevant external research
- Saved context
- The linked Atlas meeting and working note
- Access limit
- Unavailable evidence remains an open question
Separate a finding from an inference.
A brief should show what the sources support, what the interpretation is, and where evidence is still missing. Conflicting sources should stay visible.
The available feedback identifies confusion at the setup step.
- Interpretation
- The instructions may need a clearer first action.
- Open question
- Does the same friction appear beyond this sample?
Save the brief where the next task can use it.
Keep the synthesis in a note. Turn the remaining question into a task, or use the reviewed findings as the source for a presentation.
Atlas onboarding — review brief
- Note
- Findings, references, and uncertainties
- To-do
- Validate the proposed setup change
- Presentation
- Explain the decision to the team
Illustrative sources and conclusions. Research quality depends on the evidence available; a broad search is not proof that every source was covered.
What is an AI research assistant?
An AI research assistant helps investigate a question across sources, compare what they establish, and organize the result. Ghost combines public-source research with the work context you choose to bring in. Its deep research workflow keeps the query plan, evidence, and important gaps available for review—not just the final answer.
A clear question before a wider search.
Broad requests need a plan. Ghost breaks an investigation into focused questions, giving each one a purpose and choosing sources that can answer it. A product comparison, a technical explanation, and a check on recent developments should not all use the same search.
Tell Ghost what decision you are trying to make, which entities matter, and the time period to cover. Relevant terms can narrow the evidence; exclusions can keep similarly named companies or unrelated topics out of the investigation.
If the question grew out of a discussion, bring in the meeting record. Ghost can research the unresolved question while preserving what the team actually asked.
Three questions. Different evidence.
- What changed?
- Recent primary documentation and announcements.
- Does it fit the work?
- Requirements and constraints from the working brief.
- What remains uncertain?
- Missing comparisons, conflicting evidence, and questions to verify.
Deep research with the right sources.
More sources are useful only when they answer the question. Ghost’s research workflow can use the public web, community discussions, developer repositories, and other supported sources when they are relevant. A source that fails or yields thin evidence should remain a visible limitation.
Quick, standard, and deep modes change the research budget. They do not make a weak question specific, turn a search snippet into a verified fact, or guarantee that every relevant source has been found.
For information behind your own signed-in website, the AI browser agent is a separate path with its own account and permission boundaries. Public research does not silently inherit access to private pages.
For a trip, travel planning adds supported travel-option searches to the working brief. Keep your dates, constraints, and open questions together; searching for an option does not book it.
- Primary material
- Start with original documentation, official records, or other direct evidence when the question calls for it.
- Community context
- Use discussion to understand experience and questions—not as a substitute for verified product or market facts.
- Clear coverage
- Keep source failures, thin coverage, and time-window limits visible in the result.
Keep the evidence next to the conclusion.
A useful research report distinguishes what was observed from what was inferred. Ghost can assemble source-linked findings, compare competing explanations, and call out low-confidence leads rather than smoothing them into a confident narrative.
The result can be emitted as Markdown, HTML, or JSON. The format changes how you use the work; it does not remove the need to check important claims against the original sources.
When the evidence needs structured rows, the AI web scraper workflow covers permitted website data extraction, source URLs, and validation. When the reader needs a decision-ready narrative, the AI report generator workflow turns those findings into an editable report with evidence, limitations, and next steps.
For company and stock questions, use the dedicated market research workflow. Current prices and financial comparisons need appropriate structured data, not conclusions guessed from search snippets.
A recommendation you can examine.
- Finding
- State what the inspected sources support.
- Interpretation
- Explain why that finding may matter to the decision.
- Open question
- Name the missing evidence that could change the answer.
The report is the start of the next step.
Keep the useful synthesis in a research note with source links and any related meeting context. The note is an editable working record, so the brief can develop as new evidence arrives.
Ask Ghost to retrieve relevant earlier context through Memory, then verify it against the records that matter. Saved context can help frame a new investigation; it is not proof that the answer is current or complete.
Turn the approved next step into a to-do, or use the brief to build a presentation. Research, the explanation, and the action remain parts of the same work.
- Working note
- Retain the question, evidence, source links, and remaining uncertainties.
- Decision or follow-up
- Make the next action explicit and keep its supporting brief attached.
- A new deliverable
- Ask for a report or deck suited to the people who need to review it.
Product details
Research, at a glance.
Supported work, useful outputs, and what to check before you start.
- Brief
- A question, focused queries, and the source types relevant to each query. Public research and material from connected accounts use different access paths.
- Deliverable
- A report with source records, findings, and evidence gaps. Research reports can be produced as Markdown, HTML, or JSON.
- Review
- Coverage depends on what the sources return. Inspect the references and unresolved questions before using the synthesis in a decision or presentation.
One assistant. Connected work.
Ghost is a personal AI assistant for Mac. Research is a capability of the same product as Meetings, Notes, To-dos, Browser, and Memory. Bring the relevant records into the next task without treating each capability as a separate app.
Start with an unresolved question from a meeting. Research the evidence, save the synthesis in Notes, and turn the decision into a To-do or a Presentation. Ghost is the same assistant across each step.
Questions about research with Ghost.
How is deep research different from a quick web search?
A quick lookup may need only one or two sources. Deep research uses a focused query plan, source assignments, and an evidence report for a broader investigation. The selected mode changes its source budget; it does not guarantee exhaustive coverage.
Does Ghost provide sources for its research?
The research workflow preserves source records and citations with findings, along with coverage gaps and confidence information. Important claims still need checking against the original material.
Can it use earlier meetings and notes?
Yes, when you ask for work that needs them. Ghost can retrieve relevant saved context, verify the records, and bring it into a new investigation. Creating a research note does not secretly search all your past work.
Is this a scientific literature-review database?
Ghost is a personal AI assistant for work, not a replacement for a specialist scholarly database. It can investigate public sources, but should not claim access to every paper, paywalled text, or an exhaustive academic corpus.
Can I get a report instead of a chat answer?
Yes. The deep research workflow supports Markdown, HTML, and JSON output. You can also ask Ghost to keep a useful synthesis in a note or use it as the brief for another deliverable.
How do I get access?
Ghost is in private beta for macOS. Request access from the homepage. Available sources and workflows depend on the tools and connections configured for your account.
Carry the evidence into the work.
Research the question, inspect the sources, and keep the next step connected. Ghost is in private beta for macOS.
Request access