PPM

Python Package Manager oriented around persistent memory MCP tools for agentic workflows — Context+ pipelines, supermodeltools/cli analysis, and structured memory ingestion/retrieval.

PPM is the packaging and tool surface that sits alongside PMLL-style spatial memory. Memory payloads stay off-chain by default. On-chain commitment anchoring (32-byte hashes only) is live on Stellar via PMLL / pmll-anchor, not via this package surface.

Typed off-chain codework / episode payload (byte-aligned with pmll): root skill.ts. serializeCodework / hashCodework → SHA-256 → pmll-anchor store(id, commitment). Contract source and invoke docs live in drQedwards/pmll; IDs below match stellar.toml (do not invent).

Highlights

• PPM-based context stitching for agent memory and tool graphs.
• MCP tools for memory ingestion and retrieval (init, peek, set, resolve, flush, graph ops).
• MCP compat: MCPServer (mcp 2.x) with FastMCP fallback (mcp 1.x) in mcp/pmll_memory_mcp/server.py and Ppm-lib/pmll_mcp/pmll_mcp_server.py.
• Shared C core with PMLL (PMLL.h / PMLL.c, merged via PPM#70 / semantic-silo):
  • memory_silo_t — tree + slots / embed_dim (PMLL_EMBED_DIM=32) / slot_count
  • peek / peek_semantic dual-path (key|index, else cosine)
  • silo_set, free_silo, silo_embed_text, silo_cosine_similarity
  • init_pml sets assignment[i] = -1 (undecided); check_conflict treats -1 as undecided
  • SAT bridge: sat_bridge_literal / sat_bridge_clause / sat_bridge_assignment_meanings → associative memory strings
• Works with forloopcodes/contextplus hierarchical indexing and supermodeltools/cli for graphing and analysis.
• Complements drQedwards/pmll spatial memory; does not claim Stellar storage of full memory payloads.

Quick start

1. Install the memory MCP package (shared with the PMLL memory stack):

pip install pmll-memory-mcp
# or via npm: npx pmll-memory-mcp
# or: npm install -g pmll-memory-mcp

2. Register it with your MCP client (Claude Desktop example — claude_desktop_config.json):

{
  "mcpServers": {
    "pmll-memory-mcp": {
      "command": "npx",
      "args": ["pmll-memory-mcp"]
    }
  }
}

(If installed via pip you can use "command": "pmll-memory-mcp" instead.)

3. Restart the client / start a fresh session. Call init once at the start of a task, then peek before expensive operations.

Codework pickup (skill.ts → Soroban)

Off-chain only in this repo. Hash with skill.ts, then admin store on the verified pmll-anchor ID (contract lives in pmll):

import {
  exampleSemanticSiloCodework,
  hashCodework,
  toStoreArgs,
  PMLL_ANCHOR,
} from "./skill.ts";

const hash = await hashCodework(exampleSemanticSiloCodework({ skill: "ppm" }));
const { id, commitment } = toStoreArgs(hash);
// invoke pmll-anchor store -- see pmll SKILL.md / pmll-anchor/helper

Soroban ABI unchanged: init / store(id, commitment) / get / bump only. No invented fields.

Related

• PMLL spatial memory skill: https://github.com/drQedwards/pmll/blob/main/SKILL.md
• Typed payload (mirrored): skill.ts
• This repo: https://github.com/drQedwards/ppm
• Live on-chain commitments (32-byte hashes only) via pmll-anchor:
  • mainnet CCF3B64AXLS4OLY5RN4H4K2CFZAYNZCJQY5MKCKCVAKMZNH7G7F7XUUF
  • testnet CDLQR24LLFWXTNGGJVJCRXAF3ZRDWFZRUFTDZ5SJOT2J33CS7DDYP7IU
  • not part of the PPM package surface
• Network metadata: stellar.toml