> ## Documentation Index
> Fetch the complete documentation index at: https://docs.negentro.tech/llms.txt
> Use this file to discover all available pages before exploring further.

# Release Notes

> Track PiyAPI platform updates, SDK releases, and infrastructure improvements across highlights, SDKs, tools, and platform changes.

Platform updates, infrastructure improvements, and SDK releases.

<Note>
  Sample content - illustrative release notes for demonstration purposes.
</Note>

<Update label="2026-08-24" tags={["Highlight"]} description="August 2026 · Agent Plugin">
  ## PiyAPI Coding Agent Plugin launches for VS Code

  The VS Code plugin provides persistent project memory directly within the editor. The extension surfaces prior decisions, architectural conventions, and working context across isolated coding sessions.

  * **Automatic memory**: Captures working context in the background without manual tagging.
  * **Scoped recall**: Restricts `memory.search` to the current `project_id` to prevent cross-repo leakage.
  * **Inline suggestions**: Injects relevant past fixes directly into the diff view.

  <Tip>
    Developer preview. Available to design partners; general availability targeted for Q4.
  </Tip>
</Update>

<Update label="2026-08-19" description="August 2026 · SDK">
  ## Python SDK v4.2 - batched writes and async memory client

  A new `AsyncMemoryClient` and batched write support designed for high-throughput Python applications.

  * **Batch API**: Write up to 500 memories per request, reducing round trips by 90%.
  * **AsyncClient**: Mirrors the synchronous client API for `asyncio` compatibility.
</Update>

<Update label="2026-08-14" description="August 2026 · Memory">
  ## Project vs. personal memory scoping

  Memory queries and writes can now be strictly scoped to prevent crossover between shared project data and private user context.

  * **Scope param**: Pass `scope=project` or `scope=user` on every write and query operation.
  * **Isolation guarantee**: Hard boundary enforced at the storage layer, keyed by `user_id` and `project_id`.

  <Warning>
    **Migration note.** Existing memories default to `scope=user`. Update your endpoints if your application expects shared project context.
  </Warning>
</Update>

<Update label="2026-08-07" description="August 2026 · API">
  ## REST API v1.3 - richer filtering on /v1/memories

  The `/v1/memories` endpoint now supports compound filtering and date constraints.

  * **Filter operators**: Native support for `AND`/`OR` grouping structures.
  * **Session filter**: Restrict retrieval results to a specific `session_id` for conversation-level boundaries.
</Update>

<Update label="2026-07-29" tags={["Highlight"]} description="July 2026 · Platform">
  ## Native MCP support across the platform

  Native Model Context Protocol (MCP) support allows compatible clients to read and write memory directly to the PiyAPI engine.

  * **MCP server**: A single-line installation exposes native memory tools to the host environment.
  * **Tool discovery**: Hosts automatically detect `search`, `add`, and `update` tools.
</Update>

<Update label="2026-07-22" description="July 2026 · SDK">
  ## TypeScript SDK v3.0 - full type-safe memory client

  The SDK has been entirely rewritten with strict TypeScript definitions, tree-shakeable exports, and Edge runtime support.

  * **Edge-ready**: Executes in Vercel Edge and Cloudflare Workers environments with zero polyfills.
  * **Breaking change**: `client.memories.add()` now returns a strictly typed `Memory` object instead of raw JSON.
</Update>

<Update label="2026-07-16" description="July 2026 · Open Source">
  ## LangChain and LlamaIndex memory adapters

  Native memory backends are now available as drop-in replacements for the primary open-source agent frameworks.

  * **LangChain**: Implements `BaseMemory` with persistent cross-session state retention.
  * **LlamaIndex**: Distributed as a zero-config `ChatMemoryBuffer` replacement.
</Update>

<Update label="2026-07-09" description="July 2026 · Provider">
  ## UX Pilot AI and Google Gemini embedding support

  PiyAPI now supports UX Pilot AI and Google Gemini as dense vector embedding providers, alongside OpenAI.

  * **Provider config**: Configure via the `embedding_provider` project setting or explicitly per API call.
</Update>

<Update label="2026-07-02" description="July 2026 · Infrastructure">
  ## Vector retrieval latency reduced by 40%

  The core retrieval index was re-architected to reduce p95 query latency globally.

  * **HNSW tuning**: Dynamic index parameters based on real-time collection size.
  * **Result caching**: Repeated `memory.search` requests utilizing identical filters are now served directly from cache.
</Update>

<Update label="2026-06-26" tags={["Highlight"]} description="June 2026 · Product">
  ## PiyAPI Memory Platform - general availability

  The core persistent memory infrastructure is now generally available with SLA-backed uptime and strict tenant isolation.

  * **GA SLA**: 99.9% uptime commitment across all production tiers.
  * **Isolation model**: All memory records are strictly namespaced by `project_id` and `user_id`.
</Update>

<Update label="2026-06-19" description="June 2026 · Developer Tools">
  ## PiyAPI CLI 1.0

  Inspect, seed, and debug memory indices directly from the terminal.

  * **piyapi tail**: Streams live memory writes matching a specific `project_id`.
  * **piyapi seed**: Executes bulk imports from JSONL files for localized testing workflows.
</Update>

<Update label="2026-06-12" description="June 2026 · Platform">
  ## Dashboard: memory graph visualizer

  A visual interface for exploring relationships between memories, filterable by recency, category, and user entity.

  * **Graph view**: Renders topological semantic clusters directly from the production dataset.

  <Tip>
    Developer preview. Enable via the feature flags section in project settings.
  </Tip>
</Update>

<Update label="2026-06-05" description="June 2026 · Integration">
  ## Slack and Zendesk connectors

  Automatically index support conversations and team messages into scoped customer memory.

  * **Zendesk sync**: Resolved tickets are parsed and converted to memories keyed by `user_id`.
  * **Slack app**: Available via the Slack App Directory; requires channel authorization to begin indexing.
</Update>

<Update label="2026-06-01" description="June 2026 · API">
  ## REST API rate limits raised for Pro tier

  Pro workspaces receive a 5x increase to the global request ceiling across all memory endpoints.

  * **New limit**: 5,000 requests per minute per project on the REST API.
</Update>


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.