A source-grounded intelligence terminal
CrisisBrief reads the open-source picture across languages and platforms, grades every source, and writes an analyst-grade brief where each claim carries its own confidence and citation. Here is everything it does.
No credit card required · Read the feed and the CrisisBrief Index free
Define the requirement
Name a topic, time window, and source posture. CrisisBrief writes a search plan.
Collect & grade
20+ collectors sweep the open-source picture; every source is fetched and Admiralty-graded.
Extract & test
Claims are extracted, cited, scored for confidence, and checked against each other for contradictions.
Brief & review
An ICD-203 brief is drafted, graded for tradecraft, and revised, then delivered with its full ledger.
Every claim carries its evidence
Most AI tools hand you a confident paragraph and hope you trust it. CrisisBrief shows its work, line by line, source by source.
Claim-level evidence ledger
Each brief is broken into individual claims, and every claim links back to the source it came from, with the exact excerpt that supports it. Nothing is asserted that can't be traced.
Admiralty Code grading
Sources are graded for reliability (A-F) and claims for corroboration (1-6), the same NATO standard intelligence analysts use. You see how much to trust each line at a glance.
Contradiction detection
When sources disagree on dates, casualties, attribution, or severity, CrisisBrief flags the conflict instead of silently picking a side, so you know exactly where the picture is contested.
Source integrity panel
See every source behind a brief, classified by type (official government, wire service, think tank, NGO, state media), with its reliability grade. Syndicated reprints of one wire are clustered to a single origin, so ten copies of a story count once toward corroboration, not ten times.
Event timeline
Claims cluster into events and arrange chronologically, each with its confidence, supporting-claim count, and any contradictions, so you can read how a situation actually unfolded.
Evidence preservation
Every source is captured and stored when it is read, so a citation still resolves even if the original page later changes or vanishes. The evidence behind a brief does not rot.
Reads the whole open-source picture
When a crisis breaks, the signal is scattered across languages, platforms, and formats. CrisisBrief pulls from all of it in one pass.
20+ live data collectors
Web and news search, GDELT global events, ReliefWeb humanitarian alerts, Reddit, Mastodon, Telegram, Bluesky, curated RSS from governments and think tanks, Wikipedia, SEC EDGAR, and travel advisories, gathered automatically per run.
Specialist signals
NASA FIRMS fire and thermal anomalies, EIA energy-supply data, AISStream maritime vessel tracking, and YouTube, the kind of ground-truth signals that confirm or contradict the news cycle.
Source modes
Choose your evidence posture per run: Official Only for government and multilateral sources, Balanced for the mainstream picture, or Broad OSINT to sweep social and state media with stricter relevance filtering.
Multilingual, deduplicated, attributed
Material is read across English, Chinese, Russian, Arabic, Spanish, and French, deduplicated by URL, and tagged with the channel that surfaced it. Full provenance from intake to brief.
Analyst-grade reasoning, not a summary
CrisisBrief writes to the standard of a national intelligence product: structured judgments, calibrated language, and a self-review loop before anything reaches you.
ICD-203 structured briefs
Every brief leads with a BLUF (bottom line up front), then key judgments with per-claim confidence, alternative scenarios with calibrated probabilities, and clear recommendations.
Draft → review → revise
A reasoning-tier model drafts the brief, a second pass grades it for tradecraft (specificity, estimative calibration, fact-vs-analysis separation, analysis of alternatives), and a third rewrites to fix every deficiency.
Structured Analytic Techniques
Apply four formal techniques to any finished brief on demand: Key Assumptions Check, Analysis of Competing Hypotheses, Devil's Advocate, and What-If Analysis, each returning flagged findings and a bottom line.
Tailored intelligence products
Export a brief as a one-page BLUF, a SITREP (situation / assessment / outlook / recommendations), or a fuller INTSUM, each with date-time group, confidence, and citation anchors.
Red cell challenge
Before a brief is finished, an adversarial pass attacks its lead judgment and builds the strongest case that it is wrong. What survives is kept; what does not is sent back for revision.
Classification and tearline
Every brief carries a Traffic Light Protocol handling marking, and each publishes a clean, shareable tearline: the unclassified bottom line you can pass on without the workspace behind it.
Analytic continuity
Brief a topic again and CrisisBrief states plainly what changed since the last read, what is new, what firmed up, and what was walked back, so an assessment reads as a running thread, not a fresh page each time.
Collection gaps
A brief is as clear about what it could not establish as what it could: the questions left open and the reporting that would close them, so you know where the picture is thin.
See the shape of a crisis
Reading is one thing; seeing the structure is another. CrisisBrief turns claims into a map, a graph, and a single severity number.
CrisisBrief Index (CBX)
A daily 0-100 global severity score with conflict, maritime, energy, and political sub-scores, a 30-day trend, and a plain-language rationale for each day's reading. Public and free to watch.
Geospatial map
Every location mentioned in a brief is geocoded and plotted, each pin carrying its claim text, confidence, and event date. The situation at a glance, not buried in prose.
Narrative divergence
When the same event is told different ways by different camps, CrisisBrief surfaces the split instead of averaging it away, showing which sources carry which version of events.
Assessment history
Track how a judgment and its confidence move over time. Each re-brief threads into a longitudinal record, so later you can see whether the call actually held up.
A living reference, not just a feed
Every brief feeds a standing knowledge base. The actors, organisations, and places we report on accumulate into a graph, a reference library, and a searchable archive that gets richer with each run.
Entity network
The actors, organisations, and places across our published intelligence, linked wherever they share reporting. Search or filter by type, click any entity to inspect its connections and trace them without leaving the map, or switch to a sortable table to scan the whole picture.
Evergreen entity primers
A growing reference library of briefing-grade profiles on the people, organisations, and places in the holdings: who they are, why they matter, and the latest reporting on each, compiled automatically and kept current.
Entity dossiers
Pull any entity's full dossier across your holdings: every claim that mentions it, the briefs that cited them, Admiralty grades, and the other entities it moves with. All assembled from the evidence ledger, no LLM guesswork.
Ask the archive
Ask a plain-language question and get an answer synthesised from your holdings, with citations back to the underlying claims and briefs. Your accumulated reporting, queryable.
The Analyst
A conversational reasoning agent with tool access to the whole corpus: every graded claim, digest, investigation and entity dossier. Ask in plain language and it searches semantically and lexically, follows threads, builds timelines, cross-references entities and surfaces contradictions, then answers with cited evidence and ICD-203 estimative language. It holds a conversation, and tells you plainly when the holdings cannot answer. Included with Pro under a fair-use daily allowance.
Monitoring that works while you don't
The best intelligence is the warning you get before you went looking. Set your requirements once and let CrisisBrief watch.
Standing topics
Define your persistent intelligence requirements (keywords, geography, actors) and have CrisisBrief re-brief them automatically on a schedule, threading each run into a running history.
Tripwires (Indications & Warning)
Set alert conditions on your briefs: a keyword appears, confidence crosses a threshold, or a scenario's probability spikes. Get notified the moment a new brief trips one.
Footprint monitoring
Register the assets you care about (sites, suppliers, people, investments) and every brief is automatically checked for exposure, surfacing when your footprint is in the line of fire.
Forecasts & calibration
Log probability estimates on resolvable questions, then have CrisisBrief score them with Brier scores when they resolve, building a track record of how well-calibrated your judgments really are.
Burst-velocity alerts
An hourly watch compares each standing topic against its own trailing-week baseline and triggers an expedited brief when reporting spikes, so a fast-moving story reaches you off-cycle.
A standing intelligence cadence
Beyond on-demand briefs, CrisisBrief runs a published intelligence cadence you can follow without lifting a finger.
Daily, weekly & monthly briefings
Daily PDB-style updates, weekly trajectory assessments, and monthly strategic estimates, each written to the ICD-203 standard and published to the public feed.
Long-form investigations
Periodic deep-dive investigations that synthesise weeks of reporting into a single long-form analysis. The lede and key findings are free to read; the full investigation opens on Pro.
Live intelligence feed
A continuously updated situation board of the latest published assessments, each linking to a full, cited analysis you can read in seconds.
Briefings by email
Subscribe to daily, weekly, or monthly briefings delivered to your inbox, no account required, one-click unsubscribe, opt-in only.
Built to be passed up the chain
Intelligence is only useful when it reaches the person who needs it. CrisisBrief makes a finished brief easy to send, share, and present.
Export with provenance
Download any brief as styled HTML or raw Markdown. Citation anchors, source links, Admiralty grades, and evidence excerpts all travel with it.
Brief sharing
Share a brief directly with a colleague by email for read-only access. They see the full assessment and its ledger without needing to own the run.
Public links
Promoted briefs get a clean, indexable public page: a stable link you can drop into a report, a deck, or a message and know it will render.
Signed, and on the record
Intelligence you can weigh means knowing who wrote it and whether they have been right before. CrisisBrief puts both in the open.
Named analyst identity
Briefs and investigations are published under a named analyst, with structured authorship metadata, not an anonymous feed. There is a person behind the assessment.
Public track record
Resolved forecasts are scored and published openly, so the platform's calibration is on the record for anyone to inspect, not merely asserted.
Sealed, Bitcoin-anchored forecasts
Every forecast is committed to a tamper-evident hash chain the moment it is made, and the chain head is timestamped to the Bitcoin blockchain via OpenTimestamps. A call cannot be edited, deleted, or backdated, and anyone can recompute the chain to check.
Verifiable forecast receipts
When a sealed call resolves, it becomes a shareable receipt showing what we predicted and what happened, alongside the cryptographic proof that the call was committed before the outcome was known. A track record you can verify, not one you have to trust.
Benchmarked against the market
Matched forecasts are scored head-to-head against prediction-market prices on the same questions, from the same information cutoff, using the Brier rule. Every matched call is shown, win or lose; the comparison price is pinned to our seal date so it cannot be cherry-picked after the fact.
Sources earn their reliability
When a forecast resolves, the outcome is attributed back to the publishers whose reporting informed it, building a source-reliability score measured from results rather than asserted up front. A signal that compounds with every resolution.
Your intelligence, in any AI tool
CrisisBrief speaks the Model Context Protocol, so your assistant can query the holdings directly. Connect Claude, ChatGPT, Cursor or your own code and work with the data where you already are.
MCP server
Connect Claude Desktop, ChatGPT, Cursor, the OpenAI API or any MCP client to CrisisBrief. Ask the archive, pull briefs, the CrisisBrief Index, digests and entity dossiers, and even trigger new briefing runs, all from your assistant and scoped to your own holdings.
Personal access tokens
Issue revocable cb_live_ tokens from your settings, each read-only by default or opted in to trigger runs. The same entitlements, visibility scope and quota as the web app apply to every call.
Stop triaging the news. Start with a brief.
Spin up your first cited, source-graded brief in minutes, free for 14 days, no card required.