Gentleman-Programming/engram v1.15.14

on GitHub

Engram v1.15.14 is a small operational release for cloud operators and Pi users: cloud push limits are now configurable for larger self-hosted projects, and the Pi docs now clearly separate native Engram memory tools from Pi's global MCP gateway.

Highlights

• Cloud push payload limit is configurable: self-hosted cloud servers can set ENGRAM_CLOUD_MAX_PUSH_BYTES when a legitimate project outgrows the default 8 MiB request limit.
• Both cloud push paths respect the same limit: chunk push and mutation push now share the configured max body size and preserve the 8 MiB default when unset.
• Oversized mutation pushes fail clearly: mutation push now returns a structured 413 response instead of relying on an opaque hardcoded limit.
• Pi MCP status docs are clearer: Pi-native Engram mem_* tools use the gentle-engram HTTP path, while error MCP: 0/N servers is Pi's global MCP gateway health and may involve third-party MCP servers such as Notion.
• Pi setup guidance preserves MCP integrations: docs now tell users to keep pi-mcp-adapter when they rely on Notion or direct MCP flows, while troubleshooting missing native Engram tools through gentle-engram first.

Contribution highlights

Thanks to the issue reports and review pressure that shaped this release:

• #316 drove the configurable cloud push payload limit for large self-hosted projects.
• #396 came from real Pi usage where Engram memory worked while the Pi footer showed error MCP: 0/3 servers; the docs now explain that split without telling users to remove pi-mcp-adapter.
• Judgment Day review caught misleading Pi troubleshooting wording before merge, especially around preserving Notion/direct MCP flows.

Pi companion notes

No new gentle-engram npm package was published in this release. The current Pi companion remains gentle-engram@0.1.7; this release updates the main Engram binary/docs and keeps Pi setup guidance aligned with that package.

Fixes included

• ENGRAM_CLOUD_MAX_PUSH_BYTES is parsed from cloud config, validated with safe bounds, and passed into server runtime.
• Cloud chunk and mutation push handlers use the configured max payload size.
• CLI help, cloud docs, architecture docs, and changelog now mention the new cloud limit.
• Pi docs now distinguish native HTTP memory tools from the MCP gateway and add troubleshooting for error MCP: 0/N servers while mem_* still works.

Changelog

• #395 fix(cloud): make push payload limit configurable
• #397 docs(pi): clarify MCP gateway status