codex, brief me on any show or topic — a cited, skimmable brief from the smartest podcasts, rendered as a clean local report you can actually click. A working brief on live data the first time you run it — not a mockup. For "brief me on No Priors" you'd see the real 2026-06-01 episode "New Beginnings: Anthropic's IPO and Nvidia," its key_insight ("Anthropic confidentially filed for IPO at a $965B valuation, $47B revenue run-rate, surpassing OpenAI"), the top featured claims with their type and confidence, an actionable-takeaway line ("Monitor Anthropic's IPO progress and valuation vs other AI labs"), and a one-click OSC-8 link straight to the episode. The HTML report renders a claim-mix donut (event vs forecast vs analysis vs data vs statement), a guest/author byline, and honest empty-state messaging if a show has no recent episodes in window. Every fact is a Synorb-cited claim with a real source_url; nothing is invented.
A single run-it-now local CLI (Node or Python, your pick) plus an optional Codex Automation workflow. You type `codex, brief me on <topic or show>` (e.g. "brief me on No Priors" or "brief me on AI infrastructure across the top podcasts"). The tool resolves the right Synorb podcast streams, pulls real Manifests (Brief.key_insight + Signal.claims + the per-claim action line + guest/author), and writes TWO outputs: (1) a skimmable terminal brief with OSC-8 hyperlinked source episodes and a unicode-sparkline claim-mix bar, and (2) a self-contained static HTML report (no server, just open the file) with a Chart.js claim-type donut, a per-episode card grid, featured "actionable takeaways," and every card linking to its real episode source_url. Add `--watch` to register a Synorb Beacon that re-runs the same brief on the Codex scheduler (daily/weekly) so it becomes a standing digest.
It nails the exact pain of the Codex audience: you can't keep up with the smartest podcasts, and AI summaries of audio are usually hallucinated and unciteable. Brief Me is the opposite — every line traces to a real episode URL you can click in your terminal, backed by Synorb's structured claim extraction (key_insight + 15-50 atomic claims per episode + per-claim action lines that already exist in the data). It's infra-light (one file, no hosting, no GitHub), it composes Codex-native tooling (Chart.js/HTML report, OSC-8 terminal links, web-search MCP only to fill gaps with clearly-flagged non-Synorb facts), and it has a built-in retention loop via Beacons. It demos in 30 seconds on a name everyone knows (Lex, Acquired, All-In) and instantly proves Synorb's provenance moat.
No setup to learn. Codex asks a handful of normal-person questions — each with a smart default, so “you pick” always works — then builds it with live Synorb data flowing in.
Smart default: the top tech & business podcasts (Lex Fridman, Acquired, All-In, No Priors, Invest Like the Best)
Smart default: the last 30 days
Smart default: 6
Smart default: yes — open it automatically when it's done
Smart default: no — just run it once now
Editorial, magazine-grade single-column HTML report: a bold hero with the topic/show and date window, then per-episode cards each leading with the key_insight in large serif type, a guest byline, 3 featured actionable-takeaway chips, and a small claim-mix donut. Generous whitespace, warm paper background, one accent color. Feels like a hand-curated morning brief, not a data dump. Terminal mirror shows the same hierarchy with OSC-8 episode links and a one-line sparkline.
Dense, dark, operator-grade dashboard: a sticky left rail of episodes, a center claim stream where each claim is a row color-coded by claim_type (event/forecast/analysis/data/statement) with confidence badges, and a right rail of aggregate charts (claim-mix donut, claims-per-episode bars, top guests). Monospace headers, terminal-green accents — reads like a Bloomberg-for-podcasts terminal. Every claim row deep-links to the episode timestamp/source_url.
INFRA-LIGHT, no cloud hosting and no GitHub assumed. Codex is an MCP client already connected to Synorb (https://mcp.synorb.com/mcp via ~/.codex/config.toml), so at RUNTIME the product can drive live data two equivalent ways: (A) a Codex Automation/CLI workflow that calls the connected Synorb MCP tools directly, or (B) a single local script (Node or Python) that reads SYNORB_KEY from local env and calls the MCP/streamable endpoint. NEVER browser-direct-to-REST (REST CORS is a fixed allowlist). Credentials are requested locally the first time you run it (using one POST request to https://api.synorb.com/api/connect/demo-key for the key) and are never written into client code or the generated HTML. The generated HTML report is fully static with real data baked in and real episode source_url links — open it with file://, no server. For --watch, the standing digest is a saved Synorb Beacon executed by the Codex scheduler (Synorb does not cron Beacons; the platform runtime runs them by beacon_id with an explicit date window).
Credential safetySynorb credentials stay server-side or local — never in client/frontend code. You only enter them when the agent asks, with a key from synorb.com/connect/start (click Generate Free Agent Key).
Every claim, card, takeaway chip, and terminal line links to its real Synorb citation.source_url (the actual episode .mp3/show-notes URL, e.g. rss.art19.com/episodes/...mp3 for the No Priors Anthropic-IPO episode) — never an invented link. In the terminal these are OSC-8 hyperlinks so the episode title itself is clickable; in the HTML each card's title and 'Listen to the exact episode' button carry the same source_url and open in a new tab. Web-search-MCP facts (gaps only) are rendered in a visually distinct 'web context' style and labeled non-Synorb so provenance is never blurred. Honest states: if a show has no episodes in the window, coverage.status drives a clear empty/partial message instead of fabricated content.
Yes. The build prompt connects Synorb at build time to learn the real data shape, then wires live Synorb data into the running product. A working brief on live data the first time you run it — not a mockup. For "brief me on No Priors" you'd see the real 2026-06-01 episode "New Beginnings: Anthropic's IPO and Nvidia," its key_insight ("Anthropic confidentially filed for IPO at a $965B valuation, $47B revenue run-rate, surpassing OpenAI"), the top featured claims with their type and confidence, an actionable-takeaway line ("Monitor Anthropic's IPO progress and valuation vs other AI labs"), and a one-click OSC-8 link straight to the episode. The HTML report renders a claim-mix donut (event vs forecast vs analysis vs data vs statement), a guest/author byline, and honest empty-state messaging if a show has no recent episodes in window. Every fact is a Synorb-cited claim with a real source_url; nothing is invented.
Only where they are safe. INFRA-LIGHT, no cloud hosting and no GitHub assumed. Codex is an MCP client already connected to Synorb (https://mcp.synorb.com/mcp via ~/.codex/config.toml), so at RUNTIME the product can drive live data two equivalent ways: (A) a Codex Automation/CLI workflow that calls the connected Synorb MCP tools directly, or (B) a single local script (Node or Python) that reads SYNORB_KEY from local env and calls the MCP/streamable endpoint. NEVER browser-direct-to-REST (REST CORS is a fixed allowlist). Credentials are requested locally the first time you run it (using one POST request to https://api.synorb.com/api/connect/demo-key for the key) and are never written into client code or the generated HTML. The generated HTML report is fully static with real data baked in and real episode source_url links — open it with file://, no server. For --watch, the standing digest is a saved Synorb Beacon executed by the Codex scheduler (Synorb does not cron Beacons; the platform runtime runs them by beacon_id with an explicit date window).
Every claim, card, takeaway chip, and terminal line links to its real Synorb citation.source_url (the actual episode .mp3/show-notes URL, e.g. rss.art19.com/episodes/...mp3 for the No Priors Anthropic-IPO episode) — never an invented link. In the terminal these are OSC-8 hyperlinks so the episode title itself is clickable; in the HTML each card's title and 'Listen to the exact episode' button carry the same source_url and open in a new tab. Web-search-MCP facts (gaps only) are rendered in a visually distinct 'web context' style and labeled non-Synorb so provenance is never blurred. Honest states: if a show has no episodes in the window, coverage.status drives a clear empty/partial message instead of fabricated content.
A real app. Synorb is the cited data core; it also composes the platform's own tools: Codex CLI / Codex Automation workflow over the connected Synorb MCP (mcp.synorb.com/mcp via ~/.codex/config.toml), Local single-file script (Node or Python) reading SYNORB_KEY from env, Chart.js (embedded in generated static HTML for the claim-mix donut), OSC-8 terminal hyperlinks (clickable episode source_url in the terminal brief), Unicode block sparklines for the terminal claim-mix bar, Static self-contained HTML report (no hosting, opens via file://), A second MCP server for web search (gaps only, results visibly flagged as non-Synorb), Codex scheduler (runs the saved Beacon for --watch standing digests).
A few plain-English questions, each with a smart default so you can just say 'you pick': What show or topic do you want briefed?; How far back should I look for episodes?; How many episodes should the brief cover?; Do you want a clickable HTML report in addition to the terminal brief?; Should I keep this brief fresh on a schedule (a standing daily/weekly digest)?.
The same Synorb MCP connects to every major coding agent. Pick yours.
Generate credentials, connect Codex, and ship something with live, cited intelligence.