Bram 0.5.1 — shared graph workflows

Bram 0.5.1 adds graph protocol 1.1.0 / bridge 1.2.0 across the local CLI suite.

• Plan exact commands with declared input scopes and stage-specific context.
• Run approved graphs with durable checkpoints, bounded output and replay-safe retries.
• Resume only intact results whose inputs, tools, configuration and declared runtimes still match.
• Verify evidence and reject stale or tampered results.
• Generate opt-in Bash, Zsh, Fish and PowerShell completion hooks for the tools namespace.
• Keep AI diagnosis explicit and advisory; it cannot qualify checks or apply repairs.

Local checks include the full Go suite, vet and installed six-tool compatibility. Shared graph runtime recovery, locking, cancellation and freshness tests pass on native Windows 11 x64 and Debian Linux x64. See docs/GRAPH_QUALIFICATION.md for exact scope. This does not qualify every native installer or external AI provider.

Archives contain cross-built CLI binaries for macOS, Linux and Windows (amd64/arm64), documentation and BUILD.json provenance. SHA256SUMS covers the downloadable archives. macOS binaries are unsigned; these archives are not platform installers. Native make build remains the supported developer installation path.

The Go module path remains github.com/cshaiku/bram. The graph runtime is portable; launchd/mailbox installation remains macOS-specific.

Release builds use Go 1.26.9. The callable-vulnerability checks for Bram, Clyde and Goshi pass with that toolchain; module-only advisories may still be reported for unused code.

Bram 0.5.0 — shared graph workflows

Superseded by v0.5.1 for the patched Go 1.26.9 release toolchain. Use the newer release.

Bram 0.5.0 adds graph protocol 1.1.0 / bridge 1.2.0 across the local CLI suite.

• Plan exact commands with declared input scopes and stage-specific context.
• Run approved graphs with durable checkpoints, bounded output and replay-safe retries.
• Resume only intact results whose inputs, tools, configuration and declared runtimes still match.
• Verify evidence and reject stale or tampered results.
• Generate opt-in Bash, Zsh, Fish and PowerShell completion hooks for the tools namespace.
• Keep AI diagnosis explicit and advisory; it cannot qualify checks or apply repairs.

Local checks include the full Go suite, vet and installed six-tool compatibility. Shared graph runtime recovery, locking, cancellation and freshness tests pass on native Windows 11 x64 and Debian Linux x64. See docs/GRAPH_QUALIFICATION.md for exact scope. This does not qualify every native installer or external AI provider.

Archives contain cross-built CLI binaries for macOS, Linux and Windows (amd64/arm64), documentation and BUILD.json provenance. SHA256SUMS covers the downloadable archives. macOS binaries are unsigned; these archives are not platform installers. Native make build remains the supported developer installation path.

The Go module path remains github.com/cshaiku/bram. The graph runtime is portable; launchd/mailbox installation remains macOS-specific.

Bram 0.2.0 — Starlit Digital open-source release

Bram is now a Starlit Digital open-source project under 0BSD, matching Loom. It routes prompts to configured local command peers through a CLI and HTTP API.

This first public source release adds canonical module/version metadata, public setup/contribution/security docs and macOS CI. New mailbox helpers default to a home-directory path; generated agents preserve existing mailbox directories and the original service identity.

Verified locally on macOS: fresh Go tests and vet, temporary-prefix installation, mock mailbox, and isolated HTTP/CLI echo smoke. Real-provider integration, Linux/Windows qualification, authentication and resource bounds remain open. No prebuilt binaries are included.

Build from source with Go 1.24.0 or later: make compile builds only; make build installs Bram and the Grok mailbox wrapper. See the README before replacing an existing grok command.

Product: https://sltd.ca/bram/
Release evidence: https://github.com/cshaiku/bram/blob/v0.2.0/docs/RELEASE_EVIDENCE.md