DIL: Distributed Intent Ledger

DIL is durable shared memory that gives agents common vocabulary across all your knowledge domains.

A text-file-based protocol for persistent multi-agent memory across machines.

Agent-Speak: Plain Markdown, deterministic identity, scoped knowledge ownership, with anti-hallucination and proof of execution.

Human-Speak:

• DIL lets the user delegate
• DIL lets your agents receive direction, just like a human assistant would
• DIL lets your agents remember, work together, and speak your language

Quick Start

git clone https://github.com/zigmoo/distributed_intent_ledger.git $HOME/dil_agentic_memory
ln -s $HOME/dil_agentic_memory/READ_THIS_DIL_FIRST.md $HOME/dil.md

Bootstrap your agent (pick one or more):

echo 'Read $HOME/dil.md first.' >> $HOME/.claude/CLAUDE.md        # Claude Code
echo 'Read $HOME/dil.md first.' >> $HOME/.config/opencode/AGENTS.md  # OpenCode
echo 'Read $HOME/dil.md first.' >> $HOME/.codex/instructions.md      # Codex
echo 'Read $HOME/dil.md first.' >> $HOME/.pi/AGENTS.md               # Pi

That's it. Your agent reads the spec, resolves its own identity, and starts writing durable memory.

What DIL Does

• Shared memory across agents. Persistent memory across conversations. Plain Markdown on disk.
• Deterministic runtime identity resolution — agents know who and where they are without guessing.
• Scoped write boundaries — each agent owns its scope, promotes to shared only when appropriate.
• Anti-parrot execution proof — write operations must return file paths and excerpts, not just claims.
• Cross-agent task system — canonical task lifecycle with handoffs across machines and assistants.
• Mixed-model resilience — script-first, idempotent workflows that work with frontier and local models alike.

The Tacit Knowledge Problem

The core challenge in agentic engineering is not prompt engineering — it is capturing tacit knowledge.

Tacit knowledge is the millions of unwritten rules that humans absorb through apprenticeship: "don't commit without a ticket," "check who else uses this before changing it," "log what you did," "name things so the next person doesn't have to ask." Nobody teaches these in a CS degree. They're learned from code reviews, from breaking things, from watching someone senior do it right.

An agent doesn't get that apprenticeship. It gets a context window.

DIL is a tacit knowledge ledger. The preferences, policies, lessons, decisions, standards, and QC checks are all tacit knowledge made durable and retrievable. Every standard exists not because agents need special rules, but because everyone needed these rules and nobody had written them down before. The standards are for humans first, agents second — because the work is the same.

Documentation

• READ_THIS_DIL_FIRST.md — the bootstrap spec every agent reads first, every session
• Detailed README — full protocol description, repo layout, operating steps
• DIL Overview — compact human-readable summary of the whole system
• Protocol Spec v1 — normative contract (MUST/SHOULD/MAY)

Optional: Multi-Machine Sync

DIL works on one machine with zero sync infrastructure. For shared memory across machines, sync your $HOME/dil_agentic_memory/ directory with Obsidian Sync, Dropbox, or any filesystem sync tool. DIL handles identity scoping per-machine automatically.

License

Apache License 2.0 — see LICENSE and NOTICE.