Global Agent Memory v0.1.4

Global Agent Memory v0.1.4 fixes long-lived MCP connections for Claude Code and Codex.

Fixed

• Uses stateless Streamable HTTP requests so agent bridges cannot retain an expired MCP session.
• Prevents the recurring 32600: Session terminated failure after idle periods.
• Adds transport regression coverage that verifies the daemon does not issue an expiring Mcp-Session-Id.
• Documents recovery steps and clarifies that a transport failure is not an empty memory-search result.

Verification

• 78 unit tests
• 57 integration tests
• 19 contract tests
• 13 E2E tests
• 8 dashboard component tests
• Ruff, strict MyPy, TypeScript, dashboard production build, deterministic contract check
• 86.54% gated coverage
• Wheel and source distribution pass twine check

Global Agent Memory v0.1.3

Global Agent Memory v0.1.3 strengthens installation, security, recovery, testing, and public onboarding.

Highlights

• Adds the guided global-memory setup flow for Vault, daemon, dashboard, Claude Code, and Codex integration.
• Adds richer Obsidian and dashboard documentation, product visuals, comparison guidance, community templates, and contributor entry points.
• Prevents direct Markdown secrets and symlink escapes from entering the index.
• Makes backup creation and restore validation atomic and fail-closed.
• Returns structured errors for malformed dashboard JSON.
• Adds dashboard tests to make check and expands access-control, sealed-memory, stale-policy, and hostile-archive coverage.
• Modularizes the Python source package without changing the V1 CLI or MCP identifiers.

Verification

• 78 unit tests
• 57 integration tests
• 19 contract tests
• 13 E2E tests
• 8 dashboard component tests
• Ruff, strict MyPy, TypeScript, dashboard production build, deterministic contract check
• 86.54% gated coverage
• Wheel and source distribution pass twine check

See CHANGELOG.md for the complete changes.

Global Agent Memory v0.1.2

Highlights

• Refreshed the ShadCN and TypeScript dashboard across overview, access approval, memories, projects, and system views.
• Fixed the embedded-dashboard background seam by enforcing a full-height surface and matching the outer host color.
• Added managed gam-context, gam-search, gam-remember, gam-review, and gam-dashboard shortcuts for Claude Code and Codex.
• Made client registration detect broken CLI wrappers and safely use the guarded configuration fallback.

Validation

• Full make check release gate passed.
• 13 end-to-end tests passed.
• 138 unit, integration, and contract tests passed.
• 6 dashboard component tests passed.
• Ruff, strict MyPy, dashboard lint, and the production TypeScript build passed.
• Coverage: 85.20%.
• MCP contract regeneration is deterministic.
• The wheel and source archive were inspected for the dashboard and all five command skills.

Installation

Global Agent Memory is currently distributed as a source-installed local tool. See the README for installation, upgrade, dashboard, and agent integration commands.

Global Agent Memory v0.1.1

Highlights

• Added owner-controlled Protected and Sealed memory access, including exact-memory grants, permission and duration downgrades, policy enforcement, and live revocation.
• Added the ShadCN and TypeScript owner dashboard for review, search, lifecycle management, project activity, access approval, and system health.
• Fixed unsupported memory_remember types so they return a structured NOTE_INVALID response with the complete list of allowed types.
• Published the closed V1 memory-type enum through MCP discovery and CLI help.
• Scoped dashboard activity to the selected project and expanded the repository README into a complete installation, usage, architecture, security, and contributor guide.

Validation

• Full make check release gate passed.
• 13 end-to-end tests passed.
• 134 unit, integration, and contract tests passed.
• Ruff, strict MyPy, dashboard lint, and the production TypeScript build passed.
• Coverage: 85.05%.
• MCP contract regeneration is deterministic.

Installation

Global Agent Memory is currently distributed as a source-installed local tool. See the README for installation and agent integration commands.