Local Workstation

The desktop Workstation runs the full Context Lab, Agent Workspace, local data connections, runtime and investigation tools. It works without a hosted account. The current tested desktop distribution is macOS arm64, version 0.6.0. It is ad-hoc signed and not notarized, so macOS may require Open Anyway in Privacy & Security on first launch. Linux and Windows builds exist but have not completed equivalent desktop execution checks.

Download the macOS arm64 installer (0.6.0) (about 137 MB). Open the .dmg, drag SEIF Context Workstation to Applications, and launch it. On the home screen, choose Create your first AI agent for a guided agent workflow or Context engineering for exact context control. The app bundles its engine. Development from source uses the commands in the repository when access is available.

The installer is ad-hoc signed and has not been notarized. If macOS blocks the first launch, open System Settings → Privacy & Security → Open Anyway after verifying the file came from the SEIF download address.

SHA-256 of this installer: 191de9129795c45fdbf2ee566f702e99ee2ccda513d62a54a17abf286fd618a8.

Open an exported cloud project

In the cloud app, open Context engineering → Export project. Extract it and open that folder in the Workstation under Context engineering → Open project. The ZIP contains the active project contract (seif.yaml for a file-authored project or context-project.json after the Agent Workspace compiles it), Markdown/YAML knowledge, and agent-workspace.json draft definitions. It does not contain cloud sessions, published routes or credentials.

If you downloaded the JSON package instead, create a folder and extract it with the scripts/import-cloud-project.mjs helper from the Workstation source checkout:

node scripts/import-cloud-project.mjs ~/Downloads/seif-project.json ~/Documents/My-SEIF-Project

The helper validates paths, file count, size and project format before writing. A project opened locally can be committed to Git. The authoring files and agent drafts are the portable unit; cloud sessions, traces and published route state are not part of the export.

The local workflow

1. Design: add knowledge from local files, Git, inline text, HTTP or MCP sources; define consumers, objectives, layers and rules.
2. Test: preview an input and capture required/forbidden materials, status and budget as Context Tests. Tests run without a model.
3. Build: validate, run tests and create an immutable, digest-addressed Context Build.
4. Deploy: run locally, export a standalone runtime bundle or register and promote the Build through environments. Bind live provider credentials per environment; they do not enter the Build.
5. Observe: inspect Context Traces for source, rule, Build, binding and provider lineage. When a retained incident supports it, confirm an expectation and turn it into a regression guard.

The local Agent Workspace adds behavior, model connection, knowledge, capabilities, trials, evaluations, releases and application integration. Advanced workflows explains how these fit together.