ildan-memforge 0.8.0

MemForge: portable, agent-neutral persistent memory for coding agents (ILDAN).

MemForge

Typed, git-native, dynamic memory for coding agents. A markdown folder + a small spec + reference tooling. Works across Claude Code, Cursor, Aider, Codex, and GitHub Copilot Chat through thin adapters. As of v0.5, supports multi-operator teams with cryptographic attribution; as of v0.5.2, the reference CLI ships under a single memforge dispatcher with cross-platform support (macOS, Linux, native Windows); as of v0.6, query-triggered recall surfaces the memories matching a prompt on demand instead of bulk-loading the whole index.

Docs: PyPI | CHANGELOG | spec | examples

Status: pre-1.0; external PRs paused. Issues and Discussions are open. External pull requests are paused until the Contributor License Agreement infrastructure lands. See CONTRIBUTING.md. Security reports go through the private channel in SECURITY.md.

Quickstart (v0.5+, ~3 minutes)

pip install ildan-memforge
memforge --version                  # memforge 0.8.0
memforge init-operator --name "Your Name" --gen-key
cd /path/to/memory-root
memforge init-store                 # creates the signed operator-registry first
memforge recovery-init              # anchors the recovery-secret SHA256 in that registry
memforge recovery-backup-confirm --i-have-backed-up-the-secret
memforge messaging-doctor           # verify v0.5 posture: ALL CHECKS PASSED

Full quickstart at docs/quickstart.md (includes a section on commit hygiene + the signed memforge: commit-prefix grammar that audit / resolve / revoke parse). Multi-operator team bootstrap at docs/team-bootstrap.md (includes "Pick your transport: git-only or WebSocket?" decision-framing for teams sizing the v0.5 messaging adapter).

Cite this work

If you use MemForge in research or publications, please cite:

Hiltz, Mike. MemForge: Portable, agent-neutral persistent memory format for AI coding agents (v0.8.0). Zenodo, 2026. DOI: 10.5281/zenodo.20113963

The DOI above is the version-agnostic concept DOI; it always resolves to the latest archived release.

The gap MemForge fills

Coding agents have static instruction surfaces (AGENTS.md, .cursorrules, CONVENTIONS.md, .github/copilot-instructions.md, Claude Code's auto-memory) for things you want the agent to read at the start of every session. They handle the "always-true" half well.

What they don't handle: the changing half. The fact that you decided to use bun not npm last Tuesday. The reason you rejected the obvious database schema. The post-incident rule about not mocking the staging API. The seven things you've told four different agents the same way and are tired of repeating.

That state belongs in memory: typed entries that the agent can read, write, audit, and version-control alongside the code. MemForge is the format for that memory and a small set of tools that keep it healthy.

What you get

• A folder of markdown files with YAML frontmatter. Hand-editable, diffable, code-reviewable.
• Four memory types: user (facts about the developer), feedback (rules they have given), project (state about ongoing work), reference (pointers to external systems).
• A MEMORY.md index any agent loads at session start.
• Rollup subfolders for topic clusters of five or more memories, so the index stays scannable.
• Audit + integrity tooling that runs in CI: schema checks, link integrity, DLP scan, tamper-evident audit log.
• Adapters for Claude Code, Cursor, Aider, OpenAI Codex, and GitHub Copilot Chat (VS Code).

60-second install

pip install ildan-memforge

The PyPI distribution is published under ildan-memforge because the shorter memforge name is held by an unrelated project. The Python import path is still memforge, and the 20 CLI commands install with the same names (memory-audit, memory-watch, memory-dlp-scan, memforge-resolve, etc.). You can also install straight from source:

pip install git+https://github.com/ildan-ai/memforge.git

macOS / Homebrew Python (PEP 668)

On macOS with Homebrew Python, pip install against the system interpreter is blocked by PEP 668 (error: externally-managed-environment). Use one of:

# Option A: pipx (recommended; isolates the install)
brew install pipx
pipx install ildan-memforge

# Option B: dedicated venv
python3 -m venv ~/.local/share/memforge/.venv
~/.local/share/memforge/.venv/bin/pip install ildan-memforge
# then symlink the binaries into a directory on $PATH

# Option C: --user --break-system-packages (last resort; pollutes brew Python)
python3 -m pip install --user --break-system-packages ildan-memforge

The same caveat applies on most Linux distributions that mark their system Python EXTERNALLY-MANAGED (Debian 12+, Ubuntu 23.04+, Fedora 38+, etc.).

Pick a folder for your memory. The defaults the tooling assumes:

~/.claude/global-memory/                          # cross-project rules
~/.claude/projects/<USER>-claude-projects/memory/ # per-cwd state

Other layouts work too; the tooling takes a --path argument everywhere.

A memory file, end-to-end

---
name: Use bun, not npm
description: Bun is the package manager for this monorepo
type: feedback
uid: mem-2026-04-12-use-bun-not-npm
tier: index
tags: [topic:tooling, topic:javascript]
owner: web-platform
status: active
created: 2026-04-12
---

When installing or running scripts in this repo, use `bun` and `bunx`. Lockfile is `bun.lockb`.

**Why:** the workspace setup at `apps/web/package.json` relies on bun's workspace resolver; npm pulls a slightly different dependency tree and the build fails downstream.

**How to apply:** any time you reach for `npm install`, `npx`, or a `package-lock.json`, stop and use bun instead. If a script in the README still says `npm run`, treat that as a docs bug, not a directive.

That file is the unit. Every tool, every adapter, every check operates on files of exactly this shape.

When MemForge is the right tool

• You move between coding agents and want them to share the same notes.
• You want memory entries reviewable in pull requests, like code.
• You want the operator to see what the agent has written about them and edit it freely.
• You want CI gates on memory hygiene (broken links, secret leakage, schema drift).

When it is not

• Not a vector database. No embeddings, no similarity search at runtime. Use mem0 or letta if that is the shape you need.
• Not a secrets store. The DLP scanner refuses files that look like they leak credentials, but the format is human-readable plain text. Keep secrets in a vault.
• Not a runtime memory layer. MemForge is the file format and the discipline, not a service. Adapters load files into agent context at session start; nothing runs continuously.
• Not a replacement for AGENTS.md. AGENTS.md is the README for agents (static, repo-level, project setup). MemForge is the working notebook (dynamic, evolving, often per-developer). They coexist.

How it compares

	MemForge	mem0	letta	.cursorrules / AGENTS.md
Storage	Markdown files in git	Hosted vector store	Hosted agent state	Single text file
Typed entries	Yes (4 types + status)	No	Partial	No
Cross-agent	Yes (adapters)	Per-SDK	Per-SDK	Per-agent
Reviewable in PRs	Yes (diff = the file)	No (opaque store)	No	Yes
Runtime cost	Zero (load at session start)	Per-query embedding cost	Per-query	Zero
Best for	Long-lived rules + decisions	Conversational recall	Stateful agent loops	Project-level setup

The honest summary: if your agent is a long-running service that needs to remember a thousand things from yesterday's chat, you want a vector store. If your agent is a coding session that needs to remember the rules and decisions a human cares about across sessions, you want MemForge.

Architecture

flowchart LR
    Dev[Developer] -->|edits| Folder[Memory folder<br/>markdown + YAML]
    Folder -->|read at session start| AdapterCC[Claude Code adapter]
    Folder -->|read at session start| AdapterCursor[Cursor adapter]
    Folder -->|read at session start| AdapterAider[Aider adapter]
    Folder -->|read at session start| AdapterCodex[Codex adapter]
    Folder -->|read at session start| AdapterCopilot[Copilot Chat adapter]
    AdapterCC --> Agent1[Claude Code]
    AdapterCursor --> Agent2[Cursor]
    AdapterAider --> Agent3[Aider]
    AdapterCodex --> Agent4[OpenAI Codex]
    AdapterCopilot --> Agent5[GitHub Copilot Chat]
    Folder -->|CI gate| Tools[memory-audit<br/>memory-dlp-scan<br/>memory-link-rewriter<br/>memory-audit-log]
    Tools -->|fail PR on violation| GH[GitHub Actions]
    Folder -->|append-only| ChainLog[(.memforge-audit-log.jsonl<br/>tamper-evident chain)]

Adapters are thin: they translate "load this folder at session start" into the agent's native rules surface. Adding an agent means writing a new adapter, not changing the format.

Key tools

Run --help on any of them.

v0.5+ operator + identity surface (under the memforge dispatcher)

Subcommand	What it does
memforge init-operator	Generate operator-UUID + register a GPG signing key as the operator identity.
memforge init-store	Bootstrap .memforge/ in a memory-root + create a signed operator-registry.
memforge operator-registry {add|verify|remove|fresh-start}	Manage the operator-registry (add a peer operator; verify the signature; etc.).
memforge rotate-key	Generate a new long-lived key, cross-sign with the old one, lands the registry update with a 24-hour cool-down.
memforge revoke <key_id> --reason ...	Build a signed memforge: revoke <key_id> commit body.
memforge revocation-snapshot	Emit a signed snapshot of the current revocation set.
memforge memories-by-key <key_id>	List memories signed under a given key (forensic + bulk-revoke prep).
memforge revoke-memories <key_id> --bulk	Mark memories signed by a revoked key as status: superseded.
memforge upgrade-v04-memories --apply	Add v0.5 identity + signature frontmatter to v0.4 memories in-place.
memforge revoke-cache-refresh	Refresh the remote-fetch revocation cache (sparse-checkout / shallow-clone mode).
memforge messaging-doctor	Run the v0.5.1 fail-closed checklist + report posture (OK / WARN / FAIL).
memforge recovery-init	Generate ~/.memforge/recovery-secret.bin + anchor SHA256 in the signed operator-registry.
memforge recovery-backup-confirm	Acknowledge offline backup of the recovery-secret; unlocks v0.5+ writes.
memforge attest-agent	Issue a signed agent-session attestation (nonce + expires_at + capability_scope).

v0.3 / v0.4 memory-format surface (per-tool console scripts)

Tool	What it does
memory-audit	Schema + integrity check. Pass --strict in CI.
memory-audit-deep	Recursive audit: UID uniqueness, taxonomy membership, supersedes resolution, link integrity.
memory-index-gen	Build MEMORY.md from frontmatter as a CI artifact. RBAC-aware filtering for shared folders. --with-recall-index also emits the recall inverted index.
memory-cluster-suggest	Suggest rollup subfolders when a topic accumulates five or more memories.
memory-dlp-scan	Pre-commit scanner for secrets, credentials, and PII in memory bodies.
memory-link-rewriter	UID-based cross-folder link integrity (check, rename, rename-batch, upgrade). rename/rename-batch also rewrite [[wikilinks]] (including [[token|display]]) so a rename does not orphan inbound wikilinks.
memory-audit-log	Append-only tamper-evident hash chain. JSONL on disk, exports CEF for SIEM.
memory-dedup	LLM-backed near-duplicate detection. Local-only by default; reports candidates, never writes.
memory-rollup	Bulk-move primitive: create / undo / list. Maintains an undo ledger.
memory-query	Filter memories by topic, type, tag, status, owner, date range, or text.
memory-recall	Query-triggered recall (v0.6.0): inject the descriptions of memories whose triggers match a query. Reads a precompiled index; fail-open-empty.
memory-lint	Recall-readiness + token-cost quality analysis (v0.6.1). Collision-based recall scoring, always-set / description-length / body-length findings. Read-only; local-only by default, optional model-assisted suggestions behind explicit opt-in.
memory-watch	Filesystem watcher (Linux + macOS via watchdog).
memory-promote	Move a memory between folders with index fixup.
memory-frontmatter-backfill	Populate v0.3.0 frontmatter on existing memory files (migration helper).
memory-preamble-extract	Extract the MEMORY.md preamble into _memforge.yaml (Phase 1 migration helper).
agents-md-gen	Generate an AGENTS.md surface from memory frontmatter, filtered by a sensitivity ceiling + byte budget.

Competing-claim consolidation (standalone console scripts)

These ship as standalone console scripts today, not as memforge dispatcher subcommands (dispatcher integration is tracked in spec/known-limitations.md §"Reference CLI status").

Tool	What it does
memforge-resolve <topic>	Resolve a competing-claim group on a decision_topic: pick a winner, apply the superseded / superseded_by / ever_multi_member invariants, and emit a scope-locked memforge: resolve <topic> commit.
memforge-migrate-claim-block	Rewrite legacy status: to the canonical state: key inside the competing-claim fenced block of MEMORY.md.

memforge <subcommand> --help (dispatcher subcommands), <tool> --help (standalone console scripts), and tools/README.md have the long-form references.

Spec

Format invariants are in spec/SPEC.md. Topic taxonomy is in spec/taxonomy.yaml. The current package version is in pyproject.toml; the on-disk format version is in spec/VERSION. They track separately on purpose: a tooling release does not always shift the format, and a format bump is a deliberate event.

Status

Version	Date	Headline
v0.1.0	2026-04-19	First spec release.
v0.2.0	(unreleased)	Sensitivity classification (4 levels) + consumer obligations.
v0.3.0	2026-05-07	Schema expansion (uid, tier, tags, owner, last_reviewed, status), rollup-with-subfolder formalization, access labels, controlled topic taxonomy, full Phase 1 tooling, audit log, DLP scan.
v0.4.0	2026-05-08	Required-field expansion: uid, tier, tags, owner, status, created required (was optional in v0.3.x). v0.3.x files load in degraded mode. Multi-agent concurrency: five new frontmatter keys, snooze record, config file, resolve operation, canonical competing-claim block, layered Tier 1 + Tier 2 audit rule set, status enumeration BLOCKER, secure-mode adapter conformance. Sensitivity enforcement: default-on export-tier gate (memory-audit --export-tier), DLP label/content cross-check, conformance fixture set; hard floor protects privileged tier from config-disable. Closes the v0.3.x multi-user concurrency deferral.
v0.5.0	2026-05-10	Multi-identity + cryptographic attribution + WebSocket messaging adapter. Operator + agent identity as first-class peers with GPG signing. Sender-uid + sender-sequence + signed checkpoints. Signing-time-aware revocation verification + clock-skew guard. 24-hour cool-down on key rotation. Recovery-secret filesystem mode + SHA256-anchored content integrity.
v0.5.1	2026-05-10	Reference CLI ships (14 subcommands under memforge dispatcher). Closes v0.5.0 agent-session-attestation content-scope MAJOR with normative nonce + expires_at + capability_scope. Cross-cutting fail-closed posture (29-item operator reference). Privacy considerations (7 boundary statements).
v0.5.2	2026-05-10	Closes 2 BLOCKERs + 1 MAJOR from retrospective code panel. Canonical-form Unicode NFC normalization MUST on signed envelopes (closes repudiation via normalization drift). Atomic O_CREAT|O_EXCL + fsync + os.replace on write_secure_yaml + write_secure_bytes (closes TOCTOU on file create). Seen-nonce set bounding MAY -> SHOULD with explicit GC contract (closes unbounded-set DoS). Cross-platform secure-file abstraction (src/memforge/_security.py): POSIX mode bits on macOS/Linux + NTFS ACLs (via icacls) on native Windows. CI matrix extends to Ubuntu + macOS + Windows.
v0.6.0	2026-06-07	Query-triggered recall. Three optional frontmatter fields (triggers, always, do_not_inject) + the §"Recall operation" spec contract (UI-neutral; descriptions only, never bodies; fail-open-empty). New memory-recall reader + memory-index-gen --with-recall-index. All fields optional, so pre-v0.6.0 folders stay well-formed and older tools ignore the new keys.
v0.6.1 (package)	2026-06-07	Docs + packaging patch. PyPI trove classifiers added (fixes the Python-versions badge); README Status table, CLI count, and tool tables corrected; DOI badge + citation switched to the version-agnostic concept DOI. No spec change; no code change at the tag.
v0.7.0	2026-06-15	memory-lint quality CLI (recall-readiness + token-cost scoring; local-only-by-default cloud safety with sensitivity/access filtering) + memory-audit recall-field / always-set budget WARNs (spec/VERSION 0.6.0 -> 0.6.1; new §"Recall-readiness lint"). Broad security + correctness hardening: path-traversal containment (resolve/rollup/index-gen), recall/lint/dedup RBAC fail-closed, cryptographic-attribution trust root + RSA-4096 subkey verify + superseded-key refusal, revocation-snapshot floor disabled, broadened DLP coverage. Portability: IDE/OS-neutral default paths. Signed release tags + CycloneDX SBOM + a pip-audit supply-chain gate. Additive and backward-compatible.
v0.8.0	2026-06-27	memory-link-rewriter rename/rename-batch now rewrite [[wikilinks]] ([[token]] and [[token|display]]) so a rename no longer orphans inbound wikilinks: an alias-set guard (stem/name/uid) prevents false rewrites of prose or code tokens, cross-root ambiguity is skipped and listed, every rewrite is logged, and the pass is idempotent. memory-audit now warns on non-spec tier values (index/detail only; such files were previously dropped from the index with no diagnostic). Pointer-line and MEMORY.md caps raised 150 -> 180 so descriptive filename slugs leave room for the hook text (spec/VERSION 0.6.1 -> 0.6.2). Additive and backward-compatible; upgrade with pip install --upgrade ildan-memforge, no data migration required.

The reference implementation is running in production. External adoption is welcome once the CLA flow is live.

License

Apache-2.0. See LICENSE.

The "MemForge" name and logo are subject to pending trademark filings; the Apache-2.0 grant does not cover the marks. A trademark policy will land alongside the v0.3.x CLA release.

Contributing

External pull requests are paused. Issues and Discussions are open. See CONTRIBUTING.md for the current model and the post-CLA contributor flow.