Skip to content

0.6.0 - 2026-09-16

Latest
Latest

Choose a tag to compare

View all tags
@github-actions github-actions released this 17 Sep 05:46
v0.6.0
d60ec7e

Release Notes

Remote-hosting release: reach llm-orc over HTTP, serve MCP over HTTP.
No breaking changes; stdio and the local llm-orc subprocess remain the
defaults.

Added

  • plexus mcp --transport http --bind <addr> — the same tool set
    over MCP streamable HTTP at http://<bind>/mcp (default bind
    127.0.0.1:8766; --bind 127.0.0.1:0 picks a free port and logs
    listening on <url>). Many clients share one process. No auth: bind
    loopback and let a reverse proxy or the tailnet be the boundary.
    • Session semantics: each Mcp-Session-Id gets its own active
      context. set_context in one client never affects another; the
      graph, loaded specs, and lenses are shared across sessions. Sessions
      are in-memory, so a restart invalidates them and clients
      re-initialize. An unknown session id is answered with HTTP 401.
  • --llm-orc-url <base> / PLEXUS_LLM_ORC_URL on plexus mcp and
    plexus analyze — HttpClient talks to a running llm-orc serve
    (GET /health, POST /api/ensembles/{name}/execute) instead of
    spawning llm-orc locally. The flag wins over the env var; unset
    keeps SubprocessClient. No overall request timeout (long ensembles
    are normal); 10 s connect timeout; 404 → ensemble not found, 5xx →
    unavailable (extraction degrades gracefully, Invariant 47).
  • db_checkpoint (19th MCP tool + PlexusApi::db_checkpoint +
    GraphStore::checkpoint) — runs PRAGMA wal_checkpoint(TRUNCATE) and
    returns {busy, wal_pages, checkpointed_pages}. Database-level, no
    active context needed; busy: true means another connection held the
    database, retry later.
  • Library: PipelineBuilder::default_pipeline_with_client,
    PlexusMcpServer::with_client / new_session, McpServerConfig,
    McpTransport, llm_orc::{HttpClient, select_client, resolve_url}.

Shared database (ADR-017) — what a consumer may rely on

A long-lived HTTP process plus an occasional stdio process on one
SQLite file is the ADR-017 shared-DB case. The store opens in WAL mode;
every connection has a 5 s busy timeout (rusqlite default), so
concurrent writers wait rather than fail; each persist is one short
transaction (persist-per-emission, Invariant 30), so no writer holds
the lock for the length of an ingest. Readers never block on writers.
SQLite auto-checkpoints at 1000 WAL pages; db_checkpoint forces a
full truncating checkpoint on demand.

Dependencies

  • reqwest (rustls) for HttpClient; axum and the rmcp
    transport-streamable-http-server feature for the HTTP transport.

Install plexus 0.6.0

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/nrgforge/plexus/releases/download/v0.6.0/plexus-installer.sh | sh

Install prebuilt binaries via powershell script

powershell -ExecutionPolicy Bypass -c "irm https://github.com/nrgforge/plexus/releases/download/v0.6.0/plexus-installer.ps1 | iex"

Install prebuilt binaries via Homebrew

brew install nrgforge/tap/plexus

Download plexus 0.6.0

File Platform Checksum
plexus-aarch64-apple-darwin.tar.xz Apple Silicon macOS checksum
plexus-x86_64-apple-darwin.tar.xz Intel macOS checksum
plexus-x86_64-pc-windows-msvc.zip x64 Windows checksum
plexus-aarch64-unknown-linux-gnu.tar.xz ARM64 Linux checksum
plexus-x86_64-unknown-linux-gnu.tar.xz x64 Linux checksum
Assets 19
Loading