Skip to content

v2.0.3

Choose a tag to compare

@github-actions github-actions released this 11 Nov 07:16
· 453 commits to master since this release
Immutable release. Only release title and notes can be modified.

Synapse 2.0.3

A fix release on top of 2.0.2: the website chat widget works on more sites and tells you what visitors can't do, an eval run shows its Stop as soon as it starts, CLI models no longer act on a result they made up, and building from source works on macOS and Windows.

Fixed

Website chat widget

  • The widget loads on sites that hide their address. A page that asks browsers to send no referrer (with Referrer-Policy: no-referrer) used to get "null is not an allowed origin". The widget now asks for the page's origin itself, only the origin, never the path. A browser that still sends none gets a plain sentence saying so.
  • You can see which tools a visitor can't use. A tool set to ask first can't be approved by a website visitor or a caller, so it's refused for them. The tool picker for widgets and phone numbers now marks it "Asks first", and Synapse AI's widget card says which tools that affects and the two ways out: let the tool act without asking, or leave it off the widget.
  • A site keeps the scheme you typed. Asking Synapse AI for a widget on http://localhost:3000 now keeps http, and *.localhost test sites may use http too. A public site typed with http:// is refused in words, since a widget needs https there.

Evals

  • A started set shows Stop right away. After you press Run, the set reads "Starting…" until its run is seen, then shows Stop. A set that never starts says so and offers Run again, instead of quietly looking idle.

Agents and models

  • CLI models wait for a real result before acting on it. With a model through the Claude, Gemini or Copilot CLI, a reply now ends after its first call to a tool that declares it changes something outside your workspace (an API POST, for example). The model then sees that call's real result before it goes on, so it can no longer use an order number or id it guessed. Models through an API and reads in parallel are unchanged.

Under the hood

  • Building from source works on macOS and Windows. Four frontend files had names that differed from others only by capital letters, which broke the type check on case-insensitive disks. They're renamed, and a check now stops any two paths from differing only by case.
  • The store starts once, however many requests arrive first. A burst of first requests at the same moment could each try to create the database tables, and lose with "table already exists" or "database is locked". The first use is now made once and shared.

Install

macOS and Linux

curl -LsSf https://synapseorch.com/install.sh | bash

Windows (PowerShell)

irm https://synapseorch.com/install.ps1 | iex
npm, pip or Docker
npm install -g synapse-orch-ai && synapse
pip install synapse-orch-ai && synapse         # Python 3.11–3.14 and Node.js 20.9+

For Docker, see Docker: the image is synapseorchai/synapse-ai:2.0.3, also :latest.

Upgrading

  • From 2.0.x: run synapse upgrade. Your data stays where it is, and the database is backed up first.
  • From 1.x: run synapse upgrade on your 1.x install. It moves you to 2.0 the same way 2.0.0 did; read the 2.0.0 notes for what changes.

Full changelog: v2.0.2...v2.0.3 · Documentation