A transport-agnostic protocol for delivering structured intelligence payloads from publishers to registered renderers. A publisher produces one payload. Compliant gateways deliver it to any renderer. Renderers implement the protocol once.
A financial analyst receives intelligence in a dashboard. A field engineer receives it on a ruggedized device. A language model receives it as context. A product manager receives it in Slack. In each case the underlying information is the same — entities, relationships, narrative state, confidence, provenance. Only the surface differs.
Today, each delivery path is engineered independently. The dashboard team builds a REST client. The mobile team builds a push notification schema. The LLM integration team designs a context injection format. The Slack integration team writes a webhook handler. The same intelligence is formatted four times, each time with different fidelity, different update semantics, and different failure modes. There is no shared contract.
OSP defines that contract.
OSP payloads are transport-agnostic. The same envelope is valid over HTTPS webhooks, MQTT, SMS, cellular data, email, or any other transport a gateway supports. The frame header identifies the version and content type; the transport is irrelevant to the payload.
OSP does not define how intelligence is displayed. A Slack renderer formats a payload as a message block. An LLM gateway injects it as a structured system prompt fragment. A thin-client renderer displays it on an e-ink screen. The protocol is indifferent to presentation.
Payload design favors semantic density over verbosity. A well-formed OSP payload delivers actionable intelligence in the smallest possible envelope, enabling delivery to constrained environments without sacrificing meaning.
Intelligence lives at the gateway, not the device. The gateway handles tier adaptation, delivery retry, and acknowledgment tracking. The specification is published under CC BY 4.0. Any organization may implement a compatible gateway. Polari Technologies claims no proprietary position in the protocol.
OSP separates framing from payload. The frame layer defines a fixed 20-byte binary header carrying a magic version byte, flags, content type, publisher ID, monotonic sequence number, timestamp, payload length, and a CRC-16/CCITT checksum. It routes and validates — it carries no semantic content.
The payload layer defines typed schemas mapped to renderer capability tier. Core fields are present at all tiers. Extended fields — narrative state, sentiment distribution, cluster ID, geographic focus — are present at Tier 2 and above. Full graph and LLM context fields are Tier 3 only.
The flags byte encodes gateway behavior: ACK_REQUIRED for tracked delivery, EXPIRES for time-bounded payloads, DELTA for incremental updates, PRIORITY for queue-jumping, and ENCRYPTED for opaque end-to-end payloads.
Cellular SMS or equivalent. Display only. No acknowledgment required. Plain UTF-8. First entity name, truncated headline. Achievable on commodity MCU hardware with ~4KB ROM.
Low-power cellular data. Simple acknowledgment. Up to 5 entities with sentiment. MessagePack preferred for bandwidth efficiency.
HTTP/HTTPS. Full acknowledgment and retry. Adds narrative state, sentiment distribution, geographic focus, cluster ID, and topic tags to the compact baseline.
HTTP/HTTPS or MQTT. Full payload plus streaming delta updates. Entity relationship graph, narrative threads, counter-narrative detection, and LLM context fragments.
Language models have no persistent world model. Their knowledge is frozen at training time. Applications that require grounded, current intelligence must inject that intelligence into the context window at inference time.
The osp/llm-context content type packages OSP intelligence as a structured fragment suitable for direct injection. A compliant LLM gateway receives context payloads, maintains a current-state corpus keyed by cluster_id, and injects the relevant fragment before inference.
The open_questions field tells the model what it should not assert. An LLM grounded against a contested story may otherwise synthesize a confident narrative from partial evidence. Declaring open questions explicitly reduces confabulation risk.
The valid_until field enforces time-bounded grounding. A gateway delivering to an LLM integration must enforce expiry strictly — delivery of stale intelligence is worse than non-delivery.
Publishers connect intelligence pipelines to the OSP gateway and emit typed frames on events — when a story cluster forms, a trend spikes, an entity relationship changes, or a narrative state transitions. Any system that produces structured intelligence can be a publisher.
Renderers subscribe to publisher streams and receive HMAC-SHA256-signed webhook deliveries for each qualifying frame. A renderer can be a dashboard, an LLM application, a Slack integration, a mobile app, an e-ink device, or any endpoint that consumes intelligence payloads.
The whitepaper covers the complete frame specification, tier model, payload schemas, delivery and acknowledgment semantics, the publisher and renderer interfaces, LLM context infrastructure, security considerations, and a Tier 0 minimum viable implementation in pseudocode.
Independent gateways · OSP-capable renderers · Feedback from firmware engineers and protocol designers · Organizations interested in registering as publishers beyond Polari Technologies