Cost-aware agents

A low-cost web search alternative for recurring agent context.

If an agent searches the same source universe every time, the cost is not just the API call. It is repeated discovery, page fetches, cleanup, credibility checks, dedupe, citation handling, and freshness logic.

Synorb gives agents a source-grounded feed for monitored coverage, so web search becomes a fallback for gaps rather than the default path for every request.

Stop paying the agent to rediscover the same context.

Search-first loop
Repeated query-time work
Every runsearch, rank, open pages, extract, summarize, cite, dedupe
Risksnippet drift, stale pages, noisy results, missing provenance
Feed-first loop
Reusable source-grounded context
Every runquery known Streams, pull Manifests, cite source URLs, search only gaps
Riskcoverage is scoped, so the app should disclose and search outside scope when needed

A practical search budget pattern.

1

Ask Synorb what it covers

Use synorb-stream-search in MCP while building, or route production apps through REST from your backend.

2

Pull a compact Manifest set

Fetch the relevant Stream or tag scope, render source-linked rows, and store stable IDs so future runs can dedupe.

3

Search only if coverage is thin

Use web search for ad hoc facts, unknown sources, or coverage outside the monitored source universe.

Short answers.

Is this a replacement for every search API call?

No. It is a replacement for recurring search against a known source universe. Keep web search for ad hoc discovery and unknown sources.

Where should credentials live?

MCP tokens belong in your agent or MCP client config. REST api-key and secret belong only in server-side app code.

Where is pricing?

Use the current Streams pricing page for plan, cadence, and quota details. This page describes the implementation pattern.

Make search a fallback, not the whole architecture.