LCP v0.3 migration (call/quote/complete)

CONTRIBUTING.md

@@ -7,7 +7,7 @@ If you are new to Lightning, Go, or protocol work, open an issue or a draft PR a
 
 ## Quick links
 
-- Protocol draft (LCP v0.2): `docs/protocol/protocol.md`
+- Protocol draft (LCP v0.3): `docs/protocol/protocol.md`
 - Reference implementation (daemon): `go-lcpd/`
 - go-lcpd developer docs: `docs/go-lcpd/docs/`
 

README.md

@@ -5,18 +5,18 @@
 LCP (Lightning Compute Protocol) is an application-layer protocol for paying for small compute jobs over Lightning.
 It uses BOLT #1 custom messages on a direct Lightning peer connection.
 
-LCP v0.2 defines a quote → pay → stream flow:
+LCP v0.3 defines a manifest → call → quote → pay → stream → complete flow:
 - Both sides exchange `lcp_manifest` (mandatory) to advertise limits.
-- The Requester sends `lcp_quote_request`, then streams the input (`lcp_stream_begin/chunk/end` with `stream_kind=input`).
-- The Provider replies with `lcp_quote_response` containing `price_msat`, `terms_hash`, and a BOLT11 invoice.
-- After settlement, the Provider streams the result (`stream_kind=result`) and then sends `lcp_result` (terminal completion + metadata).
+- The Requester sends `lcp_call`, then streams the request payload (`lcp_stream_begin/chunk/end` with `stream_kind=request`).
+- The Provider replies with `lcp_quote` containing `price_msat`, `terms_hash`, and a BOLT11 invoice.
+- After settlement, the Provider streams the response (`stream_kind=response`) and then sends `lcp_complete` (terminal completion + metadata).
 
 LCP binds payment to job terms by setting the invoice `description_hash` to `terms_hash`.
 
 Limitations:
-- Payloads are limited by the BOLT #1 custom message size (about 65 KB), but v0.2 supports large inputs/results via chunked streams bounded by peer-declared limits (`max_payload_bytes`, `max_stream_bytes`, `max_job_bytes`).
+- Payloads are limited by the BOLT #1 custom message size (about 65 KB), but v0.3 supports large request/response bodies via chunked streams bounded by peer-declared limits (`max_payload_bytes`, `max_stream_bytes`, `max_call_bytes`).
 - The Requester and Provider must be directly peered. This leaks metadata compared to onion messages or blinded paths.
-- Payment happens before execution. This is not an atomic swap (and v0.2 also requires sending the full input stream before quoting).
+- Payment happens before execution. This is not an atomic swap (and v0.3 also requires sending the full request stream before quoting).
 
 Reference: BOLT #1 messaging (`https://github.com/lightning/bolts/blob/master/01-messaging.md`).
 
@@ -44,7 +44,7 @@ This project is unaudited. Running it against real funds and real peers can lead
 - Configuration: [docs/go-lcpd/docs/configuration.md](docs/go-lcpd/docs/configuration.md)
 - Background run + logging: [docs/go-lcpd/docs/background.md](docs/go-lcpd/docs/background.md)
 - regtest walkthrough: [docs/go-lcpd/docs/regtest.md](docs/go-lcpd/docs/regtest.md)
-- Protocol spec (LCP v0.2): [docs/protocol/protocol.md](docs/protocol/protocol.md)
+- Protocol spec (LCP v0.3): [docs/protocol/protocol.md](docs/protocol/protocol.md)
 - One-shot client (demo): [go-lcpd/tools/lcpd-oneshot](go-lcpd/tools/lcpd-oneshot)
 
 ## Docs site (Mintlify)

apps/openai-serve/README.md

@@ -87,8 +87,8 @@ High-level data flow:
 
 1. Your client calls the OpenAI-compatible HTTP endpoint (`/v1/chat/completions` or `/v1/responses`).
 2. `openai-serve` forwards the raw request body bytes as an LCP `openai.chat_completions.v1` or
-   `openai.responses.v1` task.
-3. `openai-serve` forwards the task to a local `lcpd-grpcd` (Requester) over gRPC.
+   `openai.responses.v1` method call.
+3. `openai-serve` forwards the call to a local `lcpd-grpcd` (Requester) over gRPC.
 4. The Requester talks to an LCP Provider over Lightning custom messages, requests a quote, pays, and receives the result.
 5. `openai-serve` returns the raw Provider response bytes (no re-encoding) and includes LCP metadata in response headers.
 
@@ -125,16 +125,15 @@ For a given `model`, the peer is chosen in this order:
 
 1. `OPENAI_SERVE_MODEL_MAP` (`model=peer_id;...`) if the peer is connected/LCP-ready.
 2. `OPENAI_SERVE_DEFAULT_PEER_ID` if set and connected/LCP-ready.
-3. A peer that advertises the model in `supported_tasks` (if any).
-4. Fallback to the first connected peer.
+3. A peer that advertises the required method in `supported_methods` (if any).
+4. Fallback to any connected peer.
 
 If there are no connected peers, the request fails.
 
 ### Model validation
 
 - If `OPENAI_SERVE_MODEL_ALLOWLIST` is set: `model` must be in the allowlist (unless `OPENAI_SERVE_ALLOW_UNLISTED_MODELS=true`).
-- Otherwise: if any connected peers advertise `supported_tasks`, the model must be advertised by at least one peer.
-  If no peers advertise `supported_tasks`, validation is skipped to keep the gateway usable.
+- Otherwise: validation is skipped (LCP v0.3 manifests do not advertise models; Providers enforce their own model policies).
 
 ## Safety knobs
 
@@ -147,15 +146,15 @@ If there are no connected peers, the request fails.
 This service treats logs as sensitive.
 
 - Logs MUST NOT contain raw prompts (`messages[].content`) or raw model outputs.
-- Logs include only operational metadata (e.g., peer id, job id, price, durations, and byte/token counts).
+- Logs include only operational metadata (e.g., peer id, call id, price, durations, and byte/token counts).
 - `OPENAI_SERVE_LOG_LEVEL=debug` enables more verbose request logging; keep `info` (default) for production unless needed.
 
 ## Response metadata headers
 
 `POST /v1/chat/completions` and `POST /v1/responses` include LCP metadata headers:
 
 - `X-Lcp-Peer-Id`: the chosen Provider peer id
-- `X-Lcp-Job-Id`: the job id (hex)
+- `X-Lcp-Call-Id`: the call id (hex)
 - `X-Lcp-Price-Msat`: the accepted quote price
 - `X-Lcp-Terms-Hash`: the accepted quote terms hash (hex)
 

apps/openai-serve/internal/httpapi/constants.go

@@ -1,3 +1,10 @@
 package httpapi
 
-const contentEncodingIdentity = "identity"
+const (
+	contentEncodingIdentity = "identity"
+
+	requestContentTypeJSONUTF8 = "application/json; charset=utf-8"
+
+	lcpMethodOpenAIChatCompletionsV1 = "openai.chat_completions.v1"
+	lcpMethodOpenAIResponsesV1       = "openai.responses.v1"
+)