plasmod-1.1.0

Release date: May 19, 2026

Plasmod Version | Python SDK Version | Docker Image | Node.js SDK -- | -- | -- | -- v1.1.0 | 0.1.0 (pyplasmod) | oneflybird/plasmod:v1.1 | —
Release date: May 19, 2026

Plasmod Version	Python SDK Version	Docker Image	Node.js SDK
v1.1.0	0.1.0 (pyplasmod)	oneflybird/plasmod:v1.1	—

Plasmod v1.1.0 is the first packaged preview of an agent-native database for multi-agent and LLM applications. It treats Event, Memory, State, Artifact, and Edge as first-class objects, drives state changes through an append-only WAL, and returns structured evidence on query—not only vector similarity scores. This release ships a production-oriented HTTP surface (ingest, query, admin, canonical CRUD), optional warm-segment ANN with multiple index types when built with Knowhere, batch warm retrieval with multi-agent lineage, tiered hot/warm/cold storage, and a published Docker image for local evaluation.

---

Key Features

Event-Driven Ingest & Canonical Objects

Most agent stacks store chat turns in ad hoc tables or blob stores. Provenance, session boundaries, and lifecycle state live in application code, so replay, audit, and cross-agent isolation are hard to enforce at the storage layer.

Plasmod ingests events as the single write path. Each accepted event is appended to the WAL, materialized into canonical objects (memory, state, artifact, edges), and projected into the retrieval plane. Workspace, session, and agent identifiers are part of the schema, so multi-tenant and multi-session workloads are scoped by design rather than by convention.

For more information, refer to [Ingest API (events)](https://github.com/CodeSoul-co/Plasmod/blob/main/docs/api/ingest.md) and [API Overview](https://github.com/CodeSoul-co/Plasmod/blob/main/docs/api/overview.md).

---

Structured Evidence Query

Vector-only top-K often returns similar chunks without explaining why they matter for a decision, and without tying hits back to events, versions, or graph context.

Plasmod’s query path is built for evidence-oriented responses: objects, optional graph nodes and edges, provenance, filters, and proof-style traces. Queries can be scoped by workspace and session and extended toward cold-tier retrieval when configured. This closes a common agent-memory gap where the database only ranks embeddings while the application rebuilds context and trust metadata client-side.

For more information, refer to [Query API](https://github.com/CodeSoul-co/Plasmod/blob/main/docs/api/query.md).

---

Tiered Memory Plane (Hot / Warm / Cold)

Agent memory workloads mix frequently touched session state, searchable corpora, and archived history. A single in-memory index or a single object bucket rarely fits cost, latency, and compliance together.

Plasmod routes data across hot (bounded in-memory cache), warm (segment indexes and lexical projection), and cold (S3-compatible or simulated cold store). High-salience or recent objects can sit in the hot tier; the warm tier backs ANN and hybrid retrieval; cold tier integration supports longer retention without keeping everything in RAM. Operators can point cold storage at MinIO or S3 via environment configuration in compose or custom deployments.

For more information, refer to [Architecture](https://github.com/CodeSoul-co/Plasmod/tree/main/docs/architecture) and the repository [README](https://github.com/CodeSoul-co/Plasmod/blob/main/README.md).

---

Warm Segment Vector Ingest with Pluggable ANN Index Types

Teams that already have embeddings—from batch jobs, fine-tuned encoders, or multimodal pipelines—should not be forced through an event-and-embedder path just to run ANN benchmarks or A/B index studies.

Plasmod exposes direct warm-segment ingest: callers supply vectors (and optional object IDs) and select an index type—HNSW, IVF_FLAT, IVF_PQ, IVF_SQ8, or DISKANN—with IVF tuning parameters where applicable. The server builds or rebuilds the segment index and registers object IDs for ID-resolved search. This aligns the JSON HTTP API with the existing binary ingest path so experiments can start from curl or the SDK without a custom RPC client.

Note: The published Docker image oneflybird/plasmod:v1.1 is optimized for event ingest and text query; full ANN index types require a from-source build with CGO and the Knowhere/FAISS stack.

For more information, refer to [Warm vector ingest API](https://github.com/CodeSoul-co/Plasmod/blob/main/docs/api/ingest-vectors.md).

---

Batch Warm Query with Multi-Agent Lineage

Multi-agent systems often issue **several queries in one round...