Privacy-first work continuity for DeepSeek Harness: metadata-only computer history, deterministic work episodes, and selective resume context.
The plugin will be installed here. Keep web if you are unsure.
npx -y @deepseek-ai/dsh plugin --profile web add github:ysr666/dsh-computer-history#3409862c5c5211867fd959d7543583eda6ba6e01
Compatibility and provenance
Computer History is published as dsh-computer-history and currently resolves to version 1.1.0. The Hub verifies its manifest and preserves the exact installation source for reproducible installs.
Pick up where you left off — with local, metadata-only work history for DeepSeek Harness.
Computer History turns app, resource, workspace and timing metadata into deterministic Work Episodes, so a DSH agent can understand recent work without reading the contents of your screen.
[!NOTE]
v1.0.0 is published. The same native-verified tarball passed the full installed-product journey and clean installation on Windows, macOS and Linux. npm package · GitHub Release. Host compatibility and privacy limits remain as documented.
Latest stable release: v1.1.0.npm v1.1.0 · GitHub v1.1.0 · release notes. Adds Work Memory, bounded evidence search, source-reviewed Continue, advisory Skills/Automations and privacy fixes. Three-platform packaged CI passed; native Electron screen-reader behavior and arbitrary third-party combinations are not comprehensively tested. Install on a tested DSH Host and make a consistent SQLite backup before upgrading an existing profile.
[!WARNING]
📌 Announcement — Computer History v1.1.0 stable release.
The first public Computer History release ships as one installable package for Windows, macOS and Linux. It includes native collectors, an evidence-based Timeline, Continue into a DSH session, local privacy controls and optional browser/editor companions. Published artifacts passed the three-platform packaged-client tests and the macOS end-to-end journey. Publication does not imply compatibility with untested DSH Host versions. See v1.0.0 notes, release evidence, and the verified release run. Linux still needs a real desktop X/AT-SPI session for actual capture.
Packaged DSH 0.2.0-rc.2 client with synthetic work-session data. Left: Continue, Timeline, and Work threads (example URL redacted); right: first-run privacy and recording consent.
Why this exists
A chat session knows what happened inside the chat. Your actual work often happens somewhere else: an editor, a terminal, Finder, a browser companion, or another desktop app.
Computer History is the missing continuity layer. It is built to answer questions such as:
What was I working on recently?
Which project or resource was that work attached to?
Where should I continue?
The design goal is deliberately different from productivity surveillance. The project records enough local metadata to reconstruct useful work context, while refusing to turn the screen, keyboard or document contents into a history feed.
At a glance
Computer History
Storage
Local SQLite; no remote history service required
Model
Deterministic observations → Work Episodes → selective resume context
Collection
Include-only, off by default, fail-closed on protected or uncertain surfaces
Content boundary
Metadata only — no screenshots, keystrokes, terminal output, source bodies or page bodies
DSH integration
Authenticated Host API, agent-scoped tools, History / Privacy panel
Companions
Browser and editor paths for metadata the OS cannot safely provide
The important property is what is not in that pipeline: screen pixels and document contents.
Privacy boundary
May be recorded when policy allows it
Designed never to record
application identity
screenshots or screen recordings
resource / workspace metadata
raw keyboard input or mouse coordinates
timestamps and duration evidence
clipboard contents
element role / identifier metadata
terminal output or shell history
privacy / protection state
source-file bodies
companion-provided safe resource metadata
browser page bodies
Accessibility text values or selected text
Capture starts off, application access is include-only, protected surfaces fail closed, and the Host re-screens metadata before storage. The exact guarantees, deletion model and residual risks live in SECURITY.md and docs/threat-model.md.
Packaged install + default collector handshake validated; live Notepad → UIA → Host acceptance also green
Linux
AT-SPI
Packaged install + collector handshake validated; a real desktop session still needs its X/AT-SPI bus and permissions
The shared collector contract is continuously checked on GitHub Actions across macOS, Windows and Linux. Live evidence and known limits are recorded in docs/validation-three-platforms.md.
[!IMPORTANT]
The published v1.0.0 is a single three-platform artifact, not a macOS-only package. The release pipeline builds each
native collector on its own OS, records its source commit and SHA-256, assembles all three into one plugin tarball,
then clean-installs that same tarball on macOS, Windows and Linux without a collectorExecutable override.
The installed client now also passes real first-run, Settings and successful History/Privacy browser checks on
all three platforms (measured evidence); macOS additionally passed a 56/56 full lifecycle
journey. npm OIDC Trusted Publishing and byte-for-byte npm/GitHub artifact identity were verified in the successful release.
For the verified DSH 0.2.0-rc.2 Host and an existing Web profile:
npx @deepseek-ai/dsh@0.2.0-rc.2 plugin --profile web add dsh-computer-history@1.0.0
Use the profile that your Host actually loads. web is an example, not a global install. A new profile containing only the DSH base bundle might not serve a Web UI. Do not mix bundle-managed installation with manually inserted legacy cordis.patch.yml plugin entries. Reload/restart the Host after the initial installation.
2. Allow recording deliberately
Open Computer History, choose Start recording, then open Settings → Computer History to allow the specific applications you want to record. Complete any OS accessibility permissions. Recording is off by default, the app list is include-only, and disallowed apps do not appear in the Timeline. macOS needs Accessibility access, Windows uses UI Automation, and Linux needs a real desktop X/AT-SPI session and permissions.
3. Review the Timeline and Continue
Use an allowed app, then return to Computer History → Timeline. Select a Work Episode to inspect its evidence. Choose Continue to bring bounded, evidence-backed context into a new DSH session — it does not reopen apps or files.
Use the same profile. Specify an actual published package version:
# Install a specific version (change the version for future upgrades)
npx @deepseek-ai/dsh@0.2.0-rc.2 plugin --profile web add dsh-computer-history@1.0.0
# Remove from that profile
npx @deepseek-ai/dsh@0.2.0-rc.2 plugin --profile web remove dsh-computer-history
The tested DSH Host baseline is 0.2.0-rc.2; 0.2.1-alpha.2 has additionally passed a macOS installed-product journey. This does not establish compatibility with every DSH version or platform. See release evidence and the v1.1.0 release notes.
Development from source
For development or contribution, use the source checkout:
Requirements: Node.js ^22.19.0 or >=24.0.0. Source development uses pnpm 11.7.0; native collector builds also need Swift (macOS) or Rust (Windows/Linux). See development.
Verification
pnpm verify # typecheck, lint, tests and architecture/privacy boundaries
pnpm verify:p1 # full macOS Phase 1 gate
pnpm e2e:macos # throwaway DSH Host flow on macOS
pnpm e2e:linux # Linux flow where the required desktop bus is available
pnpm benchmark:ingestion # ingestion benchmark
Pull requests run Node 22 + 24 core verification, relevant cross-platform collector checks, and the native/performance gates when the changed paths require them.
Issues and pull requests are welcome as Computer History develops. Start with CONTRIBUTING.md.
[!CAUTION]
Because this project handles sensitive local context, do not attach real history databases, pairing/session tokens, credentials, private paths, or unredacted capture logs to public issues. Use the private vulnerability-reporting path described in SECURITY.md for sensitive findings.
Acknowledgements
Computer History for DeepSeek Harness was inspired by OpenAI's Computer History, particularly its vision of helping AI assistants understand recent work context and continue where users left off. We thank OpenAI for pioneering this product direction.
We also thank the maintainers of DeepSeek Harness for building the open and extensible foundation that makes this project possible.
Computer History is an independently developed community implementation for DeepSeek Harness. It is not affiliated with or endorsed by OpenAI or DeepSeek.
