Releases: galaxycoils/darius
Releases · galaxycoils/darius
Release list
Darius v1.1.2
Functional Claude-Code-style TUI with permission gating, live events, and proper terminal interaction. 325+ tests passing.
Darius v1.1.1
Claude-Code-style TUI, web dashboard, A2A server
Darius v1.1.0
v1.1.0 — TUI, Web Dashboard, A2A & IPyKernel
New Surfaces
- TUI () — ratatui session UI with header, stream, tasks, slash commands, permission queue
- Web Dashboard () — localhost UI + SSE of UiEvent stream + POST goal
- A2A — card + tasks on the same serve listener (, )
- IPyKernel — feature-gated RLM backend ()
Architecture
- Single bus from CognitiveLoop — TUI, web SSE, A2A all subscribe
- Copper-on-near-black design tokens ( bg, accent)
- Default bind only
Tools
- , , , registered in ToolRegistry
- FTS5 full-text search (rusqlite bundled)
Tests
- 251 tests passing
- fmt clean, clippy clean
Darius v1.0.0
Changelog
All notable changes to this project will be documented in this file.
[1.0.0] - 2026-08-30
Added
- darius-memory: Single SQLite file per profile, WAL mode, FTS5 full-text search, MemoryPack capped at 3500 chars, content-hash dedupe, JSONL import/export
- darius-tools: ToolRegistry with disk spill above 32 KiB preview, TOOL line protocol, built-in memory/task tools, shell/read_file/write_file/glob tools
- darius-cognitive: CognitiveLoop (Plan → TaskBoard → ReAct → Accept), MockModel for offline tests
- Live ModelRouter:
LiveModelbehindDARIUS_LIVE_MODELenv var fordarius run - Messaging adapter: Thin I/O layer for Telegram/Discord/Slack (stub)
- Optional Jupyter/ZMQ:
IpKernelConnectionbehindrlm-ipykernelfeature - Isolation hardening:
force_terminate,detect_gvisor,terminate_with_timeout - CLI:
darius run,darius memory *,darius session-smoke - Integration e2e: Temp profile → memory → tools → cognitive loop test
Phase 2 (prior)
- Content-hash anchored PUT/CUT edits with stale-anchor rejection
- Pure-Rust RLM core with compact-safe handles
- SQLite event replay, versioned session handoffs, persistent daemon sessions
- Profile isolation, skill loading/registry, model-role routing
Known Limitations
- Live provider HTTP integration is stub (Mock works offline)
- Production messaging I/O is stub
- No Firecracker, embeddings, cloud sync, or training