Agents are getting smarter. The retrieval around them is still quite dumb. Every run starts from scratch — the same query re-described, the same sources re-picked, the same dedup re-done. For a one-off question that’s fine. For anything you actually want to stay on, it’s a full-time job.

And most of the questions worth asking are the staying-on kind: what the Fed is signaling on rates, what your competitors shipped this week, every new 8-K in a sector. You don’t ask these once — you ask them on a cadence, and you want the same well-scoped answer every time.

So we built Beacons. A Beacon is a saved retrieval recipe your agents configure once and re-run on demand. It isn’t pinned to a URL or a single website — you track a topic, an entity (a company, person, or place), or a curated cross-dimensional query. Save it once, get a beacon_id, and any agent can execute it — no re-describing the query, no maintaining your own source list, no custom dedup.

What a Beacon is

A Beacon stores the intent and the executable configuration behind a tracking goal — the Streams and source channels to draw from, the topic and entity filters, and default date-window guidance — under a stable beacon_id. It does not store content. When your agent runs a Beacon, Synorb returns fresh Manifests that match the saved configuration for the window you ask for: Signals and Briefs — structured claims and token-optimized narratives, with source provenance — not a wall of raw HTML. Your agent ingests what matters and spends fewer tokens doing it.

You dial it in once: test it over a date window, adjust the sources and filters until the results are right, then save it — so your agents reach for the version you’ve proven, not a fresh guess every run.

How it works

  1. Configure. Describe the tracking goal in plain language. Synorb’s concierge proposes Streams, source channels, and filters from real inventory — it only suggests coverage that actually exists (synorb-configure-beacon).
  2. Test & refine. Run a bounded test over a date window and see the Manifests it returns. Tighten the sources, topics, and filters until the results are right.
  3. Save. Name it, tag it, and persist the proven configuration. You get back a beacon_id (synorb-save-beacon).
  4. Execute. Any agent or backend runs synorb-manifests with the beacon_id plus a date range or lookback window, and gets matching Manifests.
  5. Manage. List, search, organize, and archive saved Beacons by tag, customer, market, or workflow (synorb-beacons, synorb-archive-beacon).

Exporting a Beacon for an agent runtime is just a handoff of the id plus the window the caller supplies on each run:

Use Beacon {beacon_id} with the Manifest tool. For this run, use the supplied date range or lookback window. Do not broaden source selection unless explicitly asked. Return matching Manifests plus the configured output format.

Run a Beacon

Running a saved Beacon is a one-liner — the beacon_id plus the window you want. Over MCP, your agent calls the same Manifest tool it already uses, and the saved scope applies automatically:

synorb-manifests beacon_id="synorb_bcn_…" lookback_hours=24

Over the REST API, pass the beacon_id on the Manifest endpoint for any Stream the Beacon covers, and supply the window yourself — Synorb doesn’t keep a “since last run” clock:

curl -H "api-key: $SYNORB_API_KEY" -H "secret: $SYNORB_API_SECRET" \ "https://api.synorb.com/streams/{stream_id}/manifests?beacon_id=synorb_bcn_…&lookback_hours=24"

Same recipe, either transport: the Beacon’s saved Streams, source channels, and filters apply automatically — you just pick the date range or lookback on each run. Run mode=orient over MCP first for a free coverage check before a billed pull.

Yours to run — runnable anywhere

It’s not a monitor that fires webhooks at you on a vendor’s schedule. You decide when it runs and over what window; Synorb keeps the recipe, the sources, and the provenance. Because a Beacon is just a beacon_id plus a date window, it drops cleanly into multi-agent harnesses, cron jobs, and your own orchestration — no vendor lock on cadence, no plumbing to maintain.

It’s not pinned to a URL either. URL watchers track an address; a Beacon tracks meaning — a topic, an entity, or a cross-dimensional query — across Synorb’s curated Streams. As Synorb adds new sources, your coverage grows without you re-pointing anything.

That turns one-off pulls into durable, named context your agents own. A research agent can save a Beacon for “AI-infrastructure developments from public-company engineering blogs and business podcasts,” hand the beacon_id to a teammate or a scheduled job, and get the same well-scoped, source-grounded context on every run — no prompt re-engineering, no drift.

What you can build

  • Competitor watch. A Beacon over a set of public-company engineering and product blogs; your agent runs it each morning and summarizes what shipped.
  • Policy monitor. A Beacon over central-bank communications; pull the last 24 hours right before a rates decision.
  • Filing queue. A Beacon over a sector’s SEC 8-K source channels; your pipeline runs it hourly and routes material events downstream.
  • Newsletter desk. A Beacon over business podcasts and investor letters on a theme; run it weekly for a sourced roundup.

Not just one kind of source

A Beacon isn’t limited to a single surface. Point it at corporate and engineering blogs, SEC filings, central-bank communications, business podcasts, investor letters, or any mix of the Streams and source channels Synorb covers. One Beacon can span several — matching a Manifest from any saved Stream while your filters narrow what comes back.

Available now

Beacons are available on every plan with per-plan count caps — Starter 10, Individual 50, Professional 250, Startup 1,000, Enterprise unlimited. The Beacon tools live on both the MCP (synorb-configure-beacon, synorb-save-beacon, synorb-beacons, synorb-archive-beacon) and the API. If you connect an agent to Synorb’s Core MCP, the Beacon tools are already on the surface. Upgrade your plan or archive an existing Beacon when you need more room.

Configure the question once. Let your agents ask it on every run — on your schedule, with full provenance.
← Product updates All posts →