@sharptrick/parley-nats

NATS JetStream backend for Parley — the network-fabric backend (plugs into a larger mesh). Implements the seam in packages/bridge-nats/src/index.ts; adding it required zero @sharptrick/parley-core changes.

Mapping

| Seam | NATS JetStream | |---|---| | topic | one JetStream stream per topic (<stream_prefix><topic>), subject <subject_prefix><topic> | | post | js.publish(subject, payload) → PubAck.seq | | cursor / backendMsgId | the stream sequence number (contiguous + monotonic per per-topic stream) | | fetchRecent({since}) | ephemeral consumer from opt_start_seq = since+1 (exclusive); no since → last-limit window | | subscribe | a consume() ordered consumer with DeliverPolicy.New — genuine events, not a poll timer | | resolveIdentity | string convention |

One stream per topic keeps sequence numbers contiguous, so the cursor is a clean per-topic monotonic integer. Core never compares cursor values — NATS delivers in seq order.

Config (backend_config)

backend_config:
  servers: "127.0.0.1:4222"   # string or list; default
  subject_prefix: "parley."    # default; topic → subject parley.<topic>
  stream_prefix: "PARLEY_"     # default; topic → stream PARLEY_<topic>
  retention_days: 30           # optional; omit to keep every message forever (the default)

Topic tokens must not contain ., *, >, or whitespace (NATS subject rules); they're sanitized to _ for safety.

Retention (optional)

retention_days sets the per-topic stream's native max_age at creation time — JetStream's own built-in retention (this plugin just supplies the value), so no separate pruning code runs here. It's off by default — max_age is unset and JetStream keeps every message forever unless you opt in. It only applies when this plugin is the one that creates the stream (the first post, fetchRecent, or subscribe on a fresh topic) — changing retention_days later does not retroactively update an already-existing stream; use nats stream edit (or recreate it) for that. As with the other backends, catch-up after the retention window just returns less history, with no error signaling that anything expired.

Multiple concurrent sessions (one backend_config per config file, same cluster)

A real deployment is several configs — one per Claude Code session plus one for the remote/chat server — all pointed at the same NATS. servers, subject_prefix/stream_prefix, and retention_days must be identical across every one of them:

• servers — the obvious one.
• subject_prefix/stream_prefix — the hidden one, same story as Redis's key_prefix: a mismatch silently maps "the same" topic to a different subject/stream while every other field still looks consistent.
• retention_days — the trickiest, because it isn't really "per config" at all: it's locked in at stream creation. Whichever session's instance is first to touch a brand-new topic wins that topic's max_age permanently; every other config's value for that topic is silently never applied. Divergence here is a race, not a choice — keep it identical everywhere so the race is harmless.

Runnable multi-config examples (two Code sessions + a remote/chat config, all sharing one cluster): examples/multi-session/nats.

Run NATS (JetStream)

Use the official nats Docker image (not authored here); JetStream is a single flag:

docker run -d --name parley-nats -p 4222:4222 nats:2.10-alpine -js

(or the maintainer dev harness: examples/dev-compose/.)

Conformance

docker run -d --name parley-nats -p 4222:4222 nats:2.10-alpine -js
npm test   # the shared @sharptrick/parley-conformance suite runs green against NATS

PARLEY_NATS_SERVERS overrides the servers; the suite skips itself if no server is reachable.