Mnemonik Blogger

A content + multi-platform publishing agent for the Mnemonik protocol. Give it a brief; it produces platform-native posts, grounds every claim in protocol facts, and publishes to Telegram, Discord, X (Twitter), and Farcaster — then records a verifiable attestation of what it published.

Mnemonik = verifiable memory for AI agents: recall it, and prove it.

Pipeline

brief ─▶ ground (Mnemonik MCP / facts) ─▶ generate (voice + guardrails)
      ─▶ render per platform (length + threads) ─▶ publish (APIs) ─▶ attest

Platforms

Platform	Transport	Threads
Telegram	Bot API → channel (bot must be admin)	yes
Discord	channel webhook (or bot token + channel id)	yes
X (Twitter)	API v2 via tweepy (paid tier for write)	yes
Farcaster	Neynar API (managed signer)	yes

Install

pip install -e ".[dev,twitter]"      # twitter extra pulls in tweepy
cp .env.example .env                  # fill in only the platforms you use

Use

# Which platforms have credentials?
mnemonik-blogger platforms

# Preview renders without posting (safe, no creds needed):
mnemonik-blogger preview \
  --title "Provable agent memory" \
  --brief "Agents forget, and you can't audit what they recalled. Mnemonik changes that." \
  --platforms auto

# Publish for real (requires credentials in env):
mnemonik-blogger post \
  --title "Provable agent memory" \
  --brief "..." \
  --link "https://mnemonik.xyz" \
  --platforms telegram,discord,x,farcaster \
  --live

By default the agent runs in dry-run (renders + reports, never posts). Real posting requires --live or MNEMONIK_DRY_RUN=false.

Grounding with the Mnemonik MCP

grounding/mnemonik.py defines a Grounding protocol. Out of the box it uses built-in facts (StaticGrounding) or a YAML file (FileGrounding, MNEMONIK_FACTS_PATH). To ground content in live, attested protocol facts and to sign each publish, wire the Mnemonik MCP (mnemonic_recall, mnemonic_sign_memory, mnemonic_verify) into a Grounding implementation — see .mcp.json and AGENTS.md.

Secrets & deployment

Nothing is hardcoded. Secrets are read from the environment as SecretStr (never logged) and injected at deploy time by a secret manager / fabric runner. See deploy/.

Develop

ruff check . && ruff format --check . && mypy && pytest -q

Status

v0.1 scaffold. Provisional Mnemonik voice/facts (derived from the protocol's MCP tool descriptions) — confirm against the official BRAND/VOICE. Live MCP grounding and the coding-fabric deploy manifest are tracked in AGENTS.md open items.