Skip to content

v1.0.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 08 Nov 12:22
v1.0.0
6ba3fc3

Signet Stable Release v1.0.0

Important

Install 1.0.1, not 1.0.0. Signet 1.0.1 fixes five security vulnerabilities in 1.0.0, in the daemon API and in connector setup. If you're on 1.0.0, run signet update install and signet daemon restart, then signet sync as described in updating an existing installation: connectors that use an API key need SIGNET_API_KEY set first. Upgrading from 0.x follows the steps below and now installs 1.0.1.

Today we’re promoting v1.0.0 to Signet’s stable channel. This is our first release with a 1.0 version number.

We started Signet expecting to build a memory server that organizations would deploy and share. Most people using it today are individual developers and heavy AI users, running several agents and switching between them depending on the task. This release is built around that: memory you run yourself, can inspect, and can carry between agents as your tools change.

The largest changes are in how memory is maintained, how context gets into Signet, how agents share it, and how you see what Signet knows. The full announcement is on the blog: Signet 1.0.

Since v0.157.3, Signet has shipped 2,869 commits across 2,736 changed files: 341,153 lines added and 194,715 lines removed. Most of that work first reached users through 500+ nightly builds.

Range: v0.157.3..v1.0.0
Previous stable: v0.157.3
New stable: v1.0.0

Upgrading from a previous release? 1.0 will not install itself through auto-update, and it upgrades your workspace layout on first start. Back up first, and read Upgrading below before you update.

Dreaming maintains your memory

The old extraction pipeline has been replaced by Dreaming, an agentic loop that maintains memory and the ontology from the evidence Signet has collected.

Extraction handled each new memory on its own. Dreaming reads new evidence alongside what Signet already knows. It consolidates related information, revisits contradictions, and proposes updates to claims. Changes are validated and recorded with citations, and the original evidence is never rewritten.

  • Claims carry real-world time: when an event happened, or the period a fact held true. That is kept separate from when Signet learned it. A claim that will change, such as a deadline or someone’s role, can be scheduled for review.
  • Passes for separate agents run in parallel. Dreaming works with local models through any OpenAI-compatible server, with configurable thinking budget, context window, and timeout.
  • The dashboard shows live traces, pass summaries, and an operation ledger.
  • Recovery from interrupted work and stalled triggers is more reliable. Duplicate passes are prevented, and passes stay within budget on large backlogs.

Sources bring in the context you already have

Sources connect notes, project documents, Notion pages, and agent conversations to Signet, with provenance.

  • Connect an Obsidian vault or Notion workspace during onboarding, or import webpages, PDFs, Office documents, and EPUBs once without an ongoing connection.
  • Conversation imports keep who said what, when, and where it came from. They resume after an interruption, and importing the same conversation twice does not duplicate its evidence.
  • Conversations can be exported as structured JSONL, filtered by agent and date, for analysis, fine-tuning preparation, or re-import.

Your context across your agents

Signet maintains integrations for Claude Code, Codex, OpenCode, OpenClaw, Hermes Agent, Pi, Oh My Pi, Kimi Code, and Muse Code. What each can do depends on the host: session context, conversation capture, and on-demand memory tools where the host supports them.

  • ChatGPT desktop connects through Signet’s native Codex plugin, with discovery and setup on macOS, Windows, and the Linux preview.
  • Kimi Code is new, with lifecycle hooks, MCP tools, and background inference through ACPX.
  • Muse Code is new on macOS and Linux, with hooks and MCP tools.
  • Hermes Agent’s native memory now takes part in Signet recall with provenance, while its memory files stay under Hermes’s control.

Agents that work together

If you run more than one agent at a time, they can now coordinate through Signet instead of through you.

  • Direct messages reach the session they’re addressed to. Idle Claude Code, Hermes Agent, and Pi sessions wake up for them; Codex reads them when its current turn ends, and other harnesses at their next prompt. Messages survive restarts and stay unread until acknowledged, or expire after seven days.
  • The message board holds topic threads that more than one session should see. Posts don’t expire, never wake anyone, and can serve as evidence for Dreaming.
  • Peer context shows which other sessions are active and what they’re working on, so an agent can pick the right one to talk to.

Better recall

  • A local cross-encoder reranks the top candidates by reading the question against each memory. It’s on by default; the model is about 90 MB and downloads on first use.
  • Keyword search matches word stems, so “deploying” finds “deployed”.
  • When Dreaming hasn’t filed a detail as a memory, recall can return a short excerpt from the original conversation, limited to what the requesting agent may read.
  • Keyword and vector results are combined by rank rather than raw score.
  • signet accelerate installs a native runtime for embeddings and reranking. signet accelerate cuda runs the reranker on an NVIDIA GPU on Linux.

Desktop, Windows, and the dashboard

  • Windows x64 is officially supported.
  • The desktop app is optional and installs with signet desktop install on macOS, Linux, and Windows x64.
  • The dashboard is redesigned around browsing memory, connecting sources and agents, changing settings, and watching Signet work. It is the same in the browser and the desktop app.
  • A memory graph maps the people, projects, and claims Signet knows about, for all agents or one at a time. Next to it, a chat answers questions about your memory with citations, using any connected model that supports tools.
  • Guided onboarding in the dashboard replaces the terminal questionnaire. Non-interactive CLI setup remains for headless installs.
  • Teams get a sign-in screen, and admins can create and revoke API keys.

Memory you can inspect

  • Recall shows where each memory came from, how it has changed, and whether it has been reviewed.
  • A claim trace returns a claim’s history, competing claims, and the exact source passages behind it, from the CLI, API, or MCP.
  • Each agent recalls only the memory it is allowed to read.
  • Text that looks like a known credential, such as an API key, is redacted from what agents see.
  • The master secret key moves to your OS keyring, and secrets stay encrypted on disk.
  • Protection status shows which parts of the workspace are backed up and flags missing or stale backups.

Reliability

  • Database work continues to move behind the database owner service, with bounded startup, shutdown, and background work, cancellation, and recovery from owner failures. The daemon recovers from crash loops on its own and backs off when the machine is busy.
  • Recall and search fixes cover scope and filter bugs, duplicate session results, caps on expensive searches, and resumable rebuilds of a damaged full-text index.
  • Imports recover from interruption, conflicting source changes are serialized, and source deletion is resumable. When the daemon is busy, transcript capture returns an error the harness can retry instead of dropping the transcript.
  • Desktop and packaging fixes cover packaged daemon startup, missing connector assets, Windows shell and path handling, macOS window controls, and AppImage updates.

Telemetry now asks first

Signet asks once before it sends anonymous usage data, in setup, signet update install, or the dashboard, and collects nothing until you say yes. Existing workspaces that never set a preference are asked too.

If you say yes, Signet sends install and version counts, feature usage, per-provider token and cost totals, and sanitized crash reports. It never sends memory content, prompts, or search queries. Every event is also written to runtime/telemetry/events.jsonl in your workspace. Change the answer with signet telemetry on or signet telemetry off, or in Settings → Advanced.

The npm package still sends one anonymous install count, version and platform only, when it installs. Set SIGNET_TELEMETRY_OPTOUT=1 before installing to skip it.

Upgrading

1.0 changes how existing installs move forward, so this part needs your attention:

  • Update manually once. Auto-update never installs a new major version. Run signet update channel stable, then signet update install, then signet daemon restart. This applies to nightly installs too.
  • Back up your workspace first. The first time the 1.0 daemon starts, it moves your workspace to the Workspace v2 layout in place. Your notes, identity files, agent.yaml, skills, and secrets stay where they are. If something blocks the move, Signet undoes it and starts on the old layout, and signet status tells you what to fix. After upgrading, don’t run older Signet versions against the workspace.
  • Turn on Dreaming. In an upgraded workspace with no Dreaming setting, Dreaming is off. The old extraction pipeline no longer runs, so no new memories are written until you set memory.dreaming.enabled: true in agent.yaml. Dreaming needs a working inference provider. New installs turn it on during setup.
  • Native installs on Linux or Windows older than v0.235.1: the old updater rejects binaries over 256 MiB, and the 1.0 binaries are larger. Rerun the native installer below instead of signet update install.
  • Keep your keyring. Once secrets move to keyring-backed storage, a workspace backup alone can’t restore them. Include the platform keychain in your recovery plan.

Installing

Choose one method; all of them install the same compiled binary.

# macOS and Linux
curl -fsSL https://signetai.sh/install.sh | bash
# Windows x64 (PowerShell)
iwr -useb https://signetai.sh/install.ps1 | iex
# npm or Bun, on any platform
npm install -g signetai
bun add -g signetai

Run signet setup after a new install. The Docker image is ghcr.io/signet-ai/signet:1.0.0.

Releases from here

Until now, stable releases were rare and most changes shipped as nightly builds, which is how the version number reached 0.240. From 1.0 on, stable releases will ship regularly under semantic versioning: patch releases for fixes, minor releases for new features, and a major release only when something breaks compatibility.

Nightly builds are paused. Installs on the nightly channel follow stable releases once they reach 1.0.0.

For the complete version-by-version history, see CHANGELOG.md.

Contributors

Thank you to the people who contributed code to this stable release:

And thank you to everyone who tested the nightly builds, reported problems, and helped get Signet to 1.0.

Assets 35
Loading