People, organizations, events, documents and posts, connected on an infinite board. A research assistant that searches, scrapes and files what it finds. Everything in one local database on your machine.
Built for investigators who deal with many actors, long timelines and a lot of evidence — and who have to explain their reasoning afterwards.
Read the documentationPan, zoom and arrange an investigation across an unbounded workspace. Group work into regions, and keep separate matters on separate boards.
Persons and organisations carry aliases, roles, affiliations and links to the evidence that supports each claim.
Every dated node is placed on an interactive timeline so sequence, gaps and overlaps are visible at a glance.
Group nodes into competing or complementary narratives and follow how each develops across the board.
Indexed search across nodes, connections and narratives, including non-Latin scripts.
Articles, social posts, documents and media are captured with metadata, archived snapshots and offline copies, so evidence survives takedowns.
A local job queue enriches new nodes in the background: profiles, article extraction, archive snapshots and media, all flagged as machine-sourced.
Generate briefing decks from the board and refine them in chat, or select nodes and compile a typeset PDF dossier: summary, chronology, network, every source.
Paste a link, a tweet, a YouTube URL, an image or plain text onto the board and the right node is created and populated automatically.
Auto-layout, regions, collapse-all, filter by node type or narrative, a minimap, and keyboard shortcuts for every node type. Large boards stay legible.
Several assistant conversations per board, auto-titled, with context compaction and a token budget so long-running cases don't run out of room.
Extend the assistant with JavaScript tools that call internal systems or partner APIs, under your own controls.
Screens from the current release, on a demo case. Pick the layout that fits the job: the board on its own, the agent on its own, or both side by side.
Each type carries the fields, previews and behaviour appropriate to it, so a board reads like a structured file rather than a wall of sticky notes. Select a type to see what it holds.
Free-form notes with Markdown formatting for headings, lists and code. The place for reasoning, working hypotheses and to-do items that should sit next to the evidence.
Dated events with precise timestamps and optional locations. Every event appears on the timeline view for sequence analysis.
Profiles for individuals: aliases, roles, affiliations and links to supporting evidence. Open reference data can enrich each profile with a description and photograph, marked as machine-sourced.
Companies, agencies and NGOs with industry, location and key personnel. Enrichment adds logos and descriptions where available.
Analytical judgements recorded with a stated confidence level and a resolution date, so assessments can be reviewed against outcomes later.
Web pages captured with title, author, publisher and body text extracted automatically, with the original link retained.
Posts from X, Reddit, Bluesky and YouTube captured with author, timestamp and engagement figures, preserved even if the source is later removed.
Images and video added by drag-and-drop or paste. Vision-capable models can describe and OCR imagery on request.
PDFs and other files attached to the board with inline previews, linked to the nodes they support.
Addresses and coordinates with map previews. Locations connect to the events and persons associated with them.
A wallet address or a single transaction on Ethereum and 50+ EVM chains, Solana or Bitcoin. Wallets carry balance, USD value, token holdings and activity; transactions are parsed into sender, recipient, value, fee, method and every decoded token, NFT or internal transfer.
A domain, IP address, email, phone number, username or URL. Right-click it, or any node that mentions one, and Pivot: WHOIS, DNS, subdomains, IP owner, email and number checks run in seconds, and you pick which results join the board as linked nodes. Each keeps the facts it was found with, and its source and date.
Threads that group related nodes and are colour-coded on the board, so competing storylines can be followed independently.
Articles ingested with automatic content extraction and an AI-generated summary, ready for relevance filtering.
Subscriptions to RSS, Reddit, Bluesky and X sources that import new items on a schedule, with keyword and relevance filters.
Labelled, colour-coded containers that organise sections of a board without altering the underlying data.
The agent panel sits alongside the canvas. It reads what is on the board, creates and links nodes, searches open sources and captures what it finds. Every tool call is shown inline as it happens, and every change can be undone.
A board has to stand up later — in a briefing, a disclosure package or a courtroom. DeepState records where each item came from, how much it is trusted and why two things are linked, so the file can be defended by someone who did not build it.
Captured sources keep the original URL, retrieval time, author and publisher, an archived copy where one exists, and offline copies of media. If the original is later removed, the node is flagged as deleted at source — the evidence remains.
Sources and claims carry a 1–5 credibility rating and a verification status — verified, unverified, disputed or debunked — with analyst notes explaining the assessment. Machine-generated content is labelled as such.
Connections are not just lines. Eleven relationship types — causes, confirms, denies, contradicts, funds, controls, employs, quotes, precedes, sources, related — each with direction, strength and a written reasoning field shown on hover.
Analytical judgements are recorded as their own nodes with a stated confidence level and a timeframe, then marked confirmed, partially confirmed or denied when the outcome is known — a running record of analytic performance.
Fifty levels of undo across node, connection and layout changes. Deletions are soft: nodes and connections are retained and can be restored. Every assistant action is listed in the transcript and reversible.
Export any board to a single .deepstate file, or share it live — over
the local network, through an end-to-end encrypted relay, or a folder your team
already controls. Every change is attributed to a device and author in an
append-only log, so a shared board is also an audit trail of who did what, when.
Ask the assistant what is new on the board since a point in time — "since yesterday", "in the last hour" — and get a structured diff of nodes, connections and narratives for shift handover or a supervisor's review.
Every source below is available as an agent tool and, where it makes sense, as a node action on the board. Results are captured to nodes with the original URL and retrieval time.
DeepState is a desktop application. There is no hosted case database, no account to create, and no requirement to send investigation material to us.
Request the security overviewFinally an investigation tool that thinks the way an analyst does — entities, timelines and evidence, not slides.
The assistant is genuinely useful. It records what it did and cites the node it took it from, which is the only way I would trust it in a file.
Local-first with real privacy. Our material never leaves the workstation, which is what made it possible to approve at all.
Maltego and i2 are built for entity graphs. Note apps are built for prose. Investigations are both — and the reading is the slow part. DeepState puts every entity, event, document and post on one canvas, and gives you an assistant that does the reading.
Right-click a domain, IP, email or phone for WHOIS, DNS, IP-owner and email lookups, run them across a selection to see what the nodes share, or tell the agent what you're after and let it chain them. You approve what lands: linked nodes, each citing its source. No transform credits.
Seventy-plus tools — web, Wikipedia, court records, SEC filings, sanctions lists, Wayback, WHOIS, social platforms, historical newspapers. It searches, scrapes, creates the nodes and cites where each one came from.
One SQLite file and a media folder on your disk. No account, no server that can read your case. Share a board with a colleague end-to-end encrypted — over your LAN, a relay that only sees ciphertext, or a USB stick.
Group findings into narratives with AI summaries, pull up the timeline, then turn the board into a slide deck or PDF the AI drafts for you. The output is the report, not a screenshot of a graph.
DeepState AI is included with every license, so it works out of the box. Or connect Claude, ChatGPT, OpenRouter, OpenCode, or a fully offline Ollama model. Switch any time; boards are unaffected.
$29 or $49 a month, per seat, cancel any time. The alternatives start in the thousands per year and assume a procurement department.
Fits alongside what you already use: import a .deepstate file from a colleague, export the board, or keep Obsidian and Maltego for what they do well.
Cancel at any time. Upgrade between plans and the difference is prorated. Not sure yet? Start with the free 14-day trial — it is the full Professional edition.
Individual researchers and freelancers
Full capability for working analysts
Teams, units and agencies
Volume rates apply from 6 seats. More than 50 seats or invoiced terms: contact sales.
Yes. Cancel at any time; access continues to the end of the billing period.
Personal plans use the AI provider you already have — Anthropic, OpenAI, OpenRouter, OpenCode Zen / Go, a Claude or ChatGPT subscription login — or a local Ollama instance. Professional and Enterprise plans additionally include managed usage through DeepState.
Yes. Enterprise administrators issue, reassign and revoke seats from the management portal at any time.
Windows, macOS (Intel and Apple silicon) and Linux. A licence is valid on any platform.
Fourteen days of the full Professional edition. No card. Your boards stay on your machine whatever you decide.