Jixu 0.5.0

Highlights

• Progressive Skills supply instructions on demand. Define a bounded catalogue with defineSkill and defineAgent({ skills }); Jixu durably records the selected body before it enters Context, and preserves it through recovery, Replay, Fork and compaction.
• The TUI now manages local Skills under /config → SKILLS. Create, edit, enable, disable and remove entries with keyboard or mouse; F2 switches configuration tabs. Apply constructs a new immutable Agent while retaining existing Thread histories.
• Unknown Tool outcomes can be resolved explicitly from the TUI. Accepted resolution and Plan receipts now describe completed coordination, and bounded Plan recovery permits ordinary execution to continue without repeating refused or exhausted controls.
• Retrieval guidance better distinguishes discovery results from page evidence. Bash timeout and cancellation terminate the foreground process tree with bounded draining and preserve indeterminate outcomes when termination cannot be confirmed.

Install

npm install -g jixu-ai@0.5.0
jixu

The package launcher requires Node.js 22.19.0 or newer. The selected standalone executable requires neither Node.js nor Bun at runtime.

Compatibility

• Native targets: Apple Silicon macOS and Linux x64 with glibc.
• Current writes use Event schema 14 and Reducer 20. Historical Event schemas 5–13 remain readable; older disposable Checkpoints are rebuilt from Events.
• Local settings use schema 7. Settings schemas 3–6 load with an empty Skills collection and preserve connection and Tool configuration.
• Changed enabled Skill metadata or bodies are incompatible with the original Agent snapshot. Restore the original enabled configuration to reopen its Thread, or create a new Thread. Disabled local entries are excluded from the Agent.
• Skills contain trusted instructions; they do not grant capabilities or Tool permissions. There is no automatic filesystem discovery, installation or hot reload.

Standalone executables

Download jixu-darwin-arm64 or jixu-linux-x64 together with SHA256SUMS and artifacts.json. These assets reuse the exact native bytes from the authoritative npm artifact set; they are not rebuilt or re-signed for GitHub Release. A downloaded executable may need chmod +x.

Safety and macOS trust

The Bash Tool is not OS-sandboxed and runs with the permissions of the Jixu process. Tool permissions control dispatch, not individual shell operations. Keep Bash on ASK unless you accept that boundary, and use a backed-up or disposable workspace.

The macOS executable is ad-hoc signed with the required JIT entitlements. It is not signed with an Apple Developer ID, is not notarized, and carries no Gatekeeper trust guarantee. Direct downloads may require manual approval in macOS.

Build evidence

Built and target-tested by Release candidate matrix 36730771438 at source commit bff7d8069860ec5cdbb8ca1863750405712f9cf7. Both native targets and the aggregated npm/pnpm/Yarn/Bun consumer gate passed before publication.

Jixu 0.4.1

Highlights

• Deterministic Tool failures now stay inside the current Agent loop. The typed failure is supplied to the next model request so the Agent can explain it, recover, or choose another action instead of ending the Thread early.
• Indeterminate Tool outcomes preserve every bounded failure reason and pending Effect. Jixu permits one explanation-only continuation, suppresses new Tool and Plan actions from that response, and then enters waiting without redispatching an unknown side effect.
• The TUI provides clearer copy feedback, and the public README plus individual npm package pages now describe the architecture, safety boundary, and install paths more directly.

Install

npm install -g jixu-ai@0.4.1
jixu

The npm package is the recommended installation path. Bun is not required at runtime.

Standalone executables

• jixu-darwin-arm64
• jixu-linux-x64

Download SHA256SUMS and artifacts.json to verify the exact target-tested bytes.

Compatibility

The supported native targets remain Apple Silicon macOS and Linux x64 with glibc. Event schemas 5 through 10 remain readable; current Threads use Event schema 11 and Reducer version 17.

Safety boundary

The Bash Tool is not OS-sandboxed and runs with the permissions of the Jixu process. Tool permissions control dispatch; they do not inspect shell commands or prevent destructive operations. Keep Bash on ASK unless you accept that risk, and use a backed-up or disposable workspace.

macOS trust disclosure

The macOS executable is ad-hoc signed with the required JIT entitlements. It is not signed with an Apple Developer ID, is not notarized, and carries no Gatekeeper trust guarantee. macOS may require additional manual approval for a directly downloaded executable.

These assets are the exact native bytes from the authoritative npm release manifest built and target-tested by Release candidate matrix 32969905715 at source commit 00921f9; they were not rebuilt or re-signed for GitHub Release.

Jixu 0.4.0

Highlights

• Jina Web Search now uses the documented /search request shape and hostname-only site constraint. Jina's exact 42206 no-results assertion is handled as a successful empty result instead of an infrastructure failure.
• A new bounded web_read Tool reads a known public HTTP(S) URL through Jina Reader with its own permission action, typed failures, durable receipt, and Replay-safe behavior.
• Existing settings keep their exact enabled Tool selection. New configurations can enable Web Read through the ordinary Tool Center and permission path.

Install

npm install -g jixu-ai@0.4.0
jixu

The npm package is the recommended installation path. Bun is not required at runtime.

Standalone executables

• jixu-darwin-arm64
• jixu-linux-x64

Download SHA256SUMS and artifacts.json to verify the exact target-tested bytes.

Compatibility

The supported native targets remain Apple Silicon macOS and Linux x64 with glibc. Event schemas 5 through 9 remain readable; current Threads use Event schema 10 and Reducer version 16.

macOS trust disclosure

The macOS executable is ad-hoc signed with the required JIT entitlements. It is not signed with an Apple Developer ID, is not notarized, and carries no Gatekeeper trust guarantee. macOS may require additional manual approval for a directly downloaded executable.

These assets are the exact native bytes from the authoritative npm release manifest built and target-tested by Release candidate matrix 32699389990 at source commit 2f53bc5; they were not rebuilt or re-signed for GitHub Release.

Jixu 0.3.0

Highlights

• Press Esc to cancel the active AI turn. Cancellation is durable, preserves public text already received, performs no automatic retry, and creates no resumable work.
• Context compilation now uses verified model capacity, source-linked Continuity Handoffs, bounded durable compaction, and an offline quality gate across both first-party provider protocols.
• The transcript shows one JIXU anchor for each uninterrupted Agent block, and Markdown tables now render as compact bordered grids with clear row and column structure.

Install

npm install -g jixu-ai@0.3.0
jixu

The npm package is the recommended installation path. Bun is not required at runtime.

Standalone executables

• jixu-darwin-arm64
• jixu-linux-x64

Download SHA256SUMS and artifacts.json to verify the exact target-tested bytes.

Compatibility

The supported native targets remain Apple Silicon macOS and Linux x64 with glibc. Event schemas 5 through 9 remain readable; current Threads use Event schema 10 and Reducer version 16.

macOS trust disclosure

The macOS executable is ad-hoc signed with the required JIT entitlements. It is not signed with an Apple Developer ID, is not notarized, and carries no Gatekeeper trust guarantee. macOS may require additional manual approval for a directly downloaded executable.

These assets are the exact native bytes from the authoritative npm release manifest built and target-tested by Release candidate matrix 32599146910 at source commit f175081; they were not rebuilt or re-signed for GitHub Release.

Jixu 0.2.0

Highlights

• Paste images directly into the Composer. Jixu stores normalized, digest-verified image Artifacts and preserves ordered text-and-image context across durable Threads, Replay, recovery, and Fork.
• Switch a Thread between Standard and Ultra with /mode. Ultra uses the strongest compatible native reasoning effort without changing the Agent instructions or adding orchestration.
• The native release boundary is now macOS arm64 and Linux x64 with glibc. Intel macOS x64 is no longer supported in this release line.

Install

npm install -g jixu-ai@0.2.0
jixu

The npm package is the recommended installation path. Bun is not required at runtime.

Standalone executables

• jixu-darwin-arm64
• jixu-linux-x64

Download SHA256SUMS and artifacts.json to verify the exact target-tested bytes.

Compatibility

Intel Mac users must pin jixu-ai@0.1.0 or move to Apple Silicon macOS or Linux x64 glibc. There is no Rosetta, runtime download, or source-build fallback in 0.2.0.

macOS trust disclosure

The macOS executable is ad-hoc signed with the required JIT entitlements. It is not signed with an Apple Developer ID, is not notarized, and carries no Gatekeeper trust guarantee. macOS may require additional manual approval for a directly downloaded executable.

These assets are the exact native bytes from the authoritative npm release manifest built and target-tested by Release candidate matrix 32488062655 at source commit c3104a3; they were not rebuilt or re-signed for GitHub Release.

Jixu 0.1.0

Jixu 0.1.0

Jixu is a durable single-Agent Harness for TypeScript. Install the TUI, run jixu, and continue work in a durable Thread.

Install

The recommended installation path is npm:

npm install -g jixu-ai
jixu

Native downloads

• jixu-darwin-arm64 — macOS on Apple silicon
• jixu-darwin-x64 — macOS on Intel
• jixu-linux-x64 — Linux on x64

After downloading the asset for your platform:

chmod +x ./jixu-<platform>
./jixu-<platform> --version

The three executables are byte-for-byte copies from the target-tested npm platform packages produced by the successful release candidate matrix at source commit e7be957. Verify them with SHA256SUMS; artifacts.json is the complete npm artifact manifest.

macOS trust notice

The macOS executables are ad-hoc signed with the required JIT entitlements. They are not signed with an Apple Developer ID and are not notarized. This release therefore does not provide Apple Gatekeeper trust guarantees, and macOS may require explicit approval before opening a downloaded executable. The current distribution policy is recorded in 360219f.

Jixu remains pre-1.0, so its public API may change before 1.0.