antaris-suite 4.5.0

Complete AI memory, routing, safety, and context management suite

Antaris Suite v4.2.0

Agent infrastructure for intelligent, secure, and memory-persistent AI systems.

v4.2.0 — Memory export/import • GCS backend stub • Hard budget enforcement • Pipeline telemetry

Packages

Package	PyPI Version	Description
antaris-memory	4.2.0	Persistent memory with BM25 search, WAL, sharding, export/import, GCS backend stub
antaris-router	4.0.1	Intelligent model routing with cost tracking, confidence gating, A/B testing
antaris-guard	4.0.1	Prompt injection detection (pattern library v1.2.0), reputation tracking, rate limiting
antaris-context	4.2.0	Context compression, hard budget enforcement, summarization, relevance scoring
antaris-pipeline	4.2.0	Agent orchestration pipeline with per-stage telemetry and OpenClaw bridge
antaris-contracts	4.0.0	Versioned state schemas, failure semantics, and debug CLI
antaris-openclaw-plugin	4.0.3	OpenClaw plugin — auto-recall, auto-ingest, Discord bridge, compaction recovery

Architecture

antaris-openclaw-plugin   (lifecycle hooks — auto-recall + auto-ingest)
        │
antaris-pipeline          (orchestration)
   ┌────┴────────────────────┐
antaris-memory         antaris-router      antaris-guard      antaris-context
(persistence)          (model selection)   (security)         (compression)

Design Principles

• Zero external dependencies on all core Python packages — stdlib only
• File-based persistence — no database required
• Multi-process safe — cross-platform FileLock using os.mkdir() atomicity
• Fully portable — no hardcoded paths, works on any machine
• Fully tested — 1,508 tests + 50-check QA runner with automated git hooks

Installation

# Install the full suite
pip install antaris-suite==4.2.0

# Or install individual packages
pip install antaris-memory antaris-guard antaris-router antaris-context antaris-pipeline

For OpenClaw plugin installation, see antaris-openclaw-plugin/INSTALL.md.

Quality Gates

Every push is gated by:

• Pre-commit hook — blocks hardcoded paths (/Users/, /home/) and require() in ESM
• Pre-push hook — runs qa_runner.py (50 checks across 4 layers), blocks push on failure
• QA Runner — python3 qa_runner.py (portability, compliance, benchmarks, bridge integration)

Set up hooks after cloning: bash scripts/install-hooks.sh

What's New in 4.2

antaris-memory — Export/Import + GCS Backend Stub

• MemorySystem.export(output_path) — Serialize all memories to a single portable JSON snapshot. Returns count of exported entries.
• MemorySystem.import_from(input_path, merge=True) — Import a snapshot into any workspace with content-hash deduplication. Returns count of newly added entries.
• GCSMemoryBackend — Google Cloud Storage backend interface fully defined and importable (from antaris_memory.backends.gcs import GCSMemoryBackend). Full cloud persistence ships in v4.3.

antaris-context — Hard Budget Enforcement

• ContextManager.render_hard_limited(budget_tokens) — Strict token ceiling: trims context to fit, raises ContextBudgetExceeded if the budget cannot be satisfied. Guarantees the returned messages never exceed the specified limit.

antaris-pipeline — Pipeline Telemetry

• PipelineTelemetry dataclass attached to every pipeline run with per-stage timing (stages: dict[str, float]), total elapsed time, and token count.
• telemetry.summary() — One-line performance report: "Pipeline: 145ms total | recall=12ms guard=5ms llm=120ms | 342 tokens".
• telemetry.slowest_stage() → (stage_name, ms) — Instantly identify pipeline bottlenecks.

antaris-router / antaris-guard

• No functional changes in 4.2. Fully compatible with the 4.2.0 suite.

Changelog

v4.2.0 (2026-02-26) — Current

• antaris-memory 4.2.0: MemorySystem.export() / import_from() with JSON snapshot and content-hash deduplication; GCS backend stub (antaris_memory.backends.gcs)
• antaris-context 4.2.0: render_hard_limited(budget_tokens) — strict token ceiling with ContextBudgetExceeded exception
• antaris-pipeline 4.2.0: PipelineTelemetry dataclass with summary() and slowest_stage() for per-run performance introspection
• antaris-router 4.0.1, antaris-guard 4.0.1 — no changes, compatible with 4.2.0 suite

v4.0.4 (2026-02-24)

• Guard TPR 20% → 100% — pattern library v1.2.0 with ~15 new patterns (DAN, ChatML, jailbreak variants, system prompt injection, harmful content, guideline bypass)
• Guard FPR 0% — zero false positives on clean business language
• Full portability audit — removed all hardcoded machine paths, ESM compatibility, __file__-relative bridge paths
• QA runner built — 4-layer automated test suite (portability, compliance, benchmarks, bridge integration)
• Git hooks — pre-commit (path check) + pre-push (QA gate) enforced on all pushes
• INSTALL.md — complete plugin installation guide with troubleshooting
• 50/50 QA score (100%)

v4.0.3 (2026-02-24)

• Bridge sys.path now __file__-relative + ANTARIS_PYTHON env var support
• Warning log when workspacePath not configured
• antaris-suite PyPI dependency pins corrected from stale v3.x → v4.x

v4.0.2 (2026-02-24)

• Fixed hardcoded /Users/moro path in plugin
• Fixed require() in ESM module
• Fixed message_in → message_received hook name
• Fixed stale api.pluginConfig reference

v4.0.1 (2026-02-24)

• antaris-pipeline 4.0.1: 8 bridge patches for v4.0.0 API compatibility
• antaris-router 4.0.1: minor fixes
• antaris-guard 4.0.1: pattern library v1.2.0

v4.0.0 (2026-02-23)

• Major version bump — all 6 packages at v4.0.0
• Component-based architecture (MemorySystem, PromptGuard, ContextManager, AdaptiveRouter)
• Bridge NDJSON protocol (stdin/stdout)
• kind: memory plugin declaration for OpenClaw memory slot
• discordChannels config for selective message ingestion
• Recovery presets (smart/minimal) for post-compaction context restoration

v3.1.0 (2026-02-21)

• Discord message bridge (auto-ingestion via message_in hook)
• Compaction hook fixes (5s timeout + 30s watchdog)
• All 6 modules consolidated under single canonical repo
• 1,508 tests passing

v3.0.0 (2026-02-20)

• Suite 3.0 launch — all 6 packages on PyPI
• 4 external reviews passed (Claude Opus, GPT-5.2, Gemini, Shiro)
• READMEs fully rewritten, docs + website updated
• 61,030× faster than mem0 (verified benchmarks)

v2.0.0 (2026-02-18)

• Suite 2.0 — 835 tests, 3 external security reviews
• OpenClaw plugin first install
• Website + docs infrastructure live

License

MIT