v0.1.14

v0.1.14 focuses on stable CLI, storage, and entity-graph behavior.

Note: harness modules, harness documentation, and harness evaluation assets in this repository remain experimental and are not part of this release's public stability guarantee.

Highlights

• Known-entity extraction now has an index-aware path that improves recall for seeded project names, internal codenames, and single-segment CamelCase entities without globally loosening first-mention extraction.
• Embedding blobs are stored as little-endian float32 values going forward, reducing embedding storage size by roughly half while preserving float64 vector APIs and cosine calculations.
• Claude Code setup now handles the $HOME/.claude collision case by writing absolute hook commands when an apparent project-local install is actually the user-global Claude config.

Added

• Entity extraction now admits wider candidate tokens when they already exist in Mnemon's known-entity index. This lets a name seeded through provided entities or earlier extraction propagate through later insights and queries without treating arbitrary new wide tokens as first-class entities.
• Store support for loading known entities from active insights. The graph engine uses this index to seed the new extraction path while continuing to ignore soft-deleted records.

Changed

• New embeddings are serialized as float32 blobs instead of float64 blobs. For common 768-dimensional embeddings, this lowers each stored embedding from 6144 bytes to 3072 bytes.
• Existing official databases migrate legacy float64 embedding blobs to float32 on open. The migration records completion in SQLite PRAGMA user_version and leaves malformed or non-legacy-looking blobs untouched rather than blocking startup.

Fixed

• Running mnemon setup for Claude Code from $HOME no longer writes relative hook commands into the user-global ~/.claude/settings.json. Mnemon now detects this collision, resolves symlinks, honors CLAUDE_CONFIG_DIR, and writes absolute hook paths so hooks work from any future session directory.
• Genuine project-local Claude Code installs continue to use relative hook commands, preserving the existing local-project behavior.

Tests and Validation

• Added vector serialization tests for float32 round trips, invalid blob lengths, and legacy float64 deserialization.
• Added store migration tests for normal legacy embeddings, malformed blobs, and non-legacy-looking blobs that should be skipped.
• Added entity extraction and store tests for known-entity propagation and soft-delete filtering.
• Release validation completed successfully:
  • go build -o mnemon .
  • make test (147 E2E checks)
  • go test ./... in a clean worktree
  • GitHub Release workflow for v0.1.14
  • GitHub master CI for commit 52fdd88

Artifacts

• macOS: mnemon_0.1.14_darwin_amd64.tar.gz, mnemon_0.1.14_darwin_arm64.tar.gz
• Linux: mnemon_0.1.14_linux_amd64.tar.gz, mnemon_0.1.14_linux_arm64.tar.gz
• Windows: mnemon_0.1.14_windows_amd64.zip
• Checksums: checksums.txt
• Homebrew tap updated via GoReleaser.

Changelog

• 9edfc4a Switch to float32 for storage (#55)
• 545f625 feat(graph): add index-aware fourth path to entity extraction
• 1982cff fix(setup): guard against project-local install colliding with the user-global Claude config (#53)

v0.1.13

v0.1.13

v0.1.13 expands Mnemon's host integration surface while keeping the stable release contract focused on the CLI, setup targets, storage behavior, and markdown/hook based agent integration. The larger harness work included in this tag remains experimental and is called out separately below.

Highlights

Hermes Agent native hook integration

This release adds mnemon setup --target hermes, a native shell-hook integration for Hermes Agent. The implementation uses Hermes' public hook surface rather than embedding Mnemon as a Hermes memory provider, so it follows Mnemon's existing model: hooks inject guidance and recalled context, while the LLM still decides what is worth storing.

The installer now:

• detects Hermes via the hermes binary or an existing ~/.hermes/ config directory
• writes ~/.hermes/skills/mnemon/SKILL.md
• installs executable hook scripts under ~/.hermes/agent-hooks/mnemon/
• patches ~/.hermes/config.yaml without removing unrelated Hermes settings
• supports mnemon setup --eject --target hermes to remove only Mnemon-owned files and hook entries

The hook mapping is:

• on_session_start records Mnemon status and guide context for the next LLM call
• pre_llm_call injects recalled memory context using Hermes' {"context": "..."} shell-hook response shape
• post_llm_call queues a lightweight remember/nudge reminder for the next turn instead of automatically writing memory
• optional on_session_finalize queues compact/finalization guidance for later context preservation

This is one implementation path for the Hermes discussion in #2. It intentionally does not close that issue by itself; a Hermes MemoryProvider/plugin path can still be evaluated independently.

Pi Coding Agent integration

v0.1.13 also includes mnemon setup --target pi, which deploys Mnemon into Pi's native extension surface.

The Pi setup target writes:

• skills/mnemon/SKILL.md
• extensions/mnemon.ts
• prompt files under the normal Mnemon prompt directory

The extension maps Mnemon lifecycle behavior onto Pi events including resources_discover, session_start, before_agent_start, agent_end, and session_before_compact. This gives Pi sessions the same core Mnemon reminders and skill discovery pattern used by the other setup targets, while keeping storage operations in the standalone mnemon binary.

Experimental harness beta

This tag includes a large experimental mnemon-harness beta surface. It is useful for internal and advanced evaluation workflows, but it is not part of the stable v0.1.13 public API contract.

Included harness work covers:

• lifecycle control-plane primitives for goals, proposals, profiles, audit records, event logs, and coordination state
• host projection and reconciliation paths for Codex and Claude Code surfaces
• eval/runtime scaffolding, replay/reporting utilities, and Codex app-server scenario assets
• daemon/job execution and trigger evaluation infrastructure
• a terminal review console for evidence, proposals, profile state, traces, hosts, and coordination views
• reorganized harness loop assets for memory, skill, eval, and goal workflows

The harness remains default-off and should not be treated as a release-path dependency for mnemon setup or other stable CLI commands.

Documentation

• English and Chinese usage docs now list the Hermes setup target.
• README setup sections now cover Hermes and Pi integration behavior.
• Harness documentation was compressed and marked as unstable beta to reduce accidental reliance on experimental contracts.

Install / Upgrade

Homebrew users can upgrade with:

brew update
brew upgrade mnemon

Direct downloads are available below for:

• macOS amd64 / arm64
• Linux amd64 / arm64
• Windows amd64

Validation

The release workflow completed successfully and published GoReleaser artifacts plus the Homebrew formula. The release test job ran the E2E suite before publishing.

Additional local checks run before tagging:

• go test ./internal/setup ./cmd
• go build -o mnemon .
• bash scripts/check_bilingual_sync.sh
• make test (147 E2E checks passed)

Notes

• GitHub Actions reported Node.js 20 deprecation warnings for upstream actions (actions/checkout, actions/setup-go, and goreleaser/goreleaser-action). The release completed successfully; this is workflow maintenance to address later.
• GoReleaser reported deprecation warnings for some config keys (archives.format, archives.format_overrides.format, and brews). The artifacts and Homebrew formula were still published successfully.

v0.1.12

v0.1.12 focuses on stable CLI, storage, and entity-graph behavior.

Note: harness modules, harness documentation, and harness evaluation assets in this repository remain experimental and are not part of this release's public stability guarantee.

Changelog

• efbbd85 fix(setup): align OpenClaw prompt path with MNEMON_DATA_DIR
• dd1d61d fix(setup): scope prompt files to MNEMON_DATA_DIR (#42)

v0.1.11

v0.1.11 improves the documentation and installed agent guidance around the mnemon import workflow introduced in v0.1.10.

Note: harness modules, harness documentation, and harness evaluation assets in this repository remain experimental and are not part of this release's public stability guarantee.

Highlights

Import guide internationalization

• Reworked docs/IMPORT.md into the English canonical import guide.
• Added docs/zh/IMPORT.md as the Chinese import guide with the same schema reference, examples, command usage, output explanation, FAQ, and LLM extraction prompt.
• Added language-switch links between the English and Chinese import guides.
• Linked the import guide from both README documentation indexes.

Setup skill guidance for import workflows

• Updated the installed mnemon skill templates for Codex, Claude Code, Nanobot, and OpenClaw.
• Added mnemon import --dry-run <file> and mnemon import <file> to each skill's command list.
• Added a short “Import Historical Chats” workflow so host agents know how to:
  • create a memory_draft.json with schema_version: "1"
  • include insight fields such as content, category, importance, tags, entities, and optional created_at
  • include optional explicit edges with source_index, target_index, edge_type, weight, and reason
  • validate with mnemon import --dry-run before writing
  • verify imports with mnemon status, mnemon search, or mnemon recall
  • check the errors field because imports can partially succeed

Validation

• go test ./...
• go build -o mnemon .
• git diff --check
• Release workflow passed make build and make test.

Links

• Full changelog: v0.1.10...v0.1.11

v0.1.10

v0.1.10 adds the first supported workflow for importing historical chat exports into Mnemon memory, plus a small Claude setup fix included since v0.1.9.

Note: harness modules, harness documentation, and harness evaluation assets in this repository remain experimental and are not part of this release's public stability guarantee.

Highlights

Historical memory import

• Added mnemon import <file> for bulk-importing memory draft JSON files into the active or selected store.
• Imported insights go through Mnemon's normal write behavior: duplicate/conflict detection, embedding generation when Ollama is available, entity/causal/semantic graph construction, lifecycle scoring, and automatic capacity pruning.
• Added mnemon import --dry-run to validate a draft without writing to SQLite.
• Added mnemon import --no-diff for trusted imports where every draft item should be inserted as new memory.

Versioned draft schema

• Added the schema_version: "1" memory draft schema under internal/importdraft.
• Draft files support top-level source, an insights array, and optional explicit edges.
• Insight fields include content, category, importance, tags, entities, per-insight source, and RFC 3339 created_at for preserving historical timestamps.
• Explicit edges use source_index, target_index, edge_type, weight, and optional reason, allowing LLM-generated imports to preserve strong relationships that heuristics may not infer.

Import documentation and LLM prompt

• Added docs/IMPORT.md with the full draft schema reference, field constraints, category guidance, importance guidance, edge-type guidance, examples, and operational notes.
• Included a copy-paste LLM extraction prompt for converting chat exports or notes into memory_draft.json.
• Updated docs/USAGE.md with the new import command examples.

Correctness fixes in the import path

• Historical imports no longer corrupt temporal ordering when importing older created_at memories into a store that already contains newer memories.
• The import path disables real-time temporal edge generation during writes, then repairs affected source timelines after all backdated insights are inserted so the imported memories are placed into the global chronological chain.
• Explicit draft edges now affect effective_importance before pruning. Import finalization inserts explicit edges, repairs temporal edges, refreshes touched insight scores, and only then runs AutoPrune.

Internal changes

• Added internal graph.EngineOptions and TemporalMode so import can skip real-time temporal edge generation without changing mnemon remember behavior.
• Added store helpers for source-ordered active insight scans and typed single-edge deletion, used by temporal repair.

Validation

• Added unit tests for draft validation, including required content, category and importance constraints, RFC 3339 timestamps, edge index bounds, and source fallback behavior.
• Added command integration tests for backdated temporal repair and effective-importance refresh after explicit draft edges.
• Release workflow passed make build and make test; local release prep also passed go test ./..., go build -o mnemon ., and git diff --check.

Also included since v0.1.9

• Fixed Claude setup to use a valid memory subagent type (#40).

Links

• Import PR: #41
• Related issue: #32
• Full changelog: v0.1.9...v0.1.10

v0.1.9

v0.1.9 makes mnemon recall lighter for routine agent use.

Changed

• BREAKING: mnemon recall now returns a compact JSON shape by default instead of the previous full debug payload.
• Default recall results now include the fields agents usually need: id, content, category, importance, intent, matched_via, confidence, and rounded score.
• Verbose/debug fields such as signals, timestamps, access_count, tags, entities, source, traversal metadata, and the full embedded insight object are omitted from the default output.

Migration

• Use mnemon recall <query> --verbose to restore the previous full payload shape.
• Scripts or SDKs that parse paths like .results[].insight.id, .results[].signals, or .meta should either add --verbose or migrate to the new compact fields.

Notes

• --basic recall output is unchanged.
• Stored data and the SQLite schema are unchanged; this is a CLI JSON output contract change only.
• Harness modules and harness evaluation assets remain experimental and are not part of this release public stability guarantee.

Changelog

• feat(recall): make default output LLM-friendly
• docs(release): prepare v0.1.9 notes

v0.1.8

v0.1.8 focuses on stable CLI, storage, and entity-graph behavior.

Note: harness modules, harness documentation, and harness evaluation assets in this repository remain experimental and are not part of this release's public stability guarantee.

Changelog

• f443aef docs(release): prepare v0.1.8 notes
• 638fdf6 feat(embed): add --embed-model CLI flag for runtime model override (#37)

v0.1.7

v0.1.7 focuses on stable CLI, storage, and entity-graph behavior.

Note: harness modules, harness documentation, and harness evaluation assets in this repository remain experimental and are not part of this release's public stability guarantee.

Changelog

• 6aa0751 docs(site): simplify lifecycle control plane model
• ced6cba feat(docs): add rendered docs navigation
• cd94385 feat: add privacy-safe memory receipts
• 4f7b827 fix: skip writes in IncrementAccessCount and LogOp when read-only

v0.1.6

v0.1.6 focuses on stable CLI, storage, and entity-graph behavior.

Note: harness modules, harness documentation, and harness evaluation assets in this repository remain experimental and are not part of this release's public stability guarantee.

Changelog

• 2269b43 chore(release): prepare v0.1.6 notes
• 7618e0d docs(contributing): centralize changelog updates at release time
• 3acdc34 feat(setup): add Codex integration
• 1e11a0d feat: Add Nanobot support
• 623426d fix: address nanobot PR review feedback

v0.1.5

v0.1.5 focuses on stable CLI, storage, and entity-graph behavior.

Note: harness modules, harness documentation, and harness evaluation assets in this repository remain experimental and are not part of this release's public stability guarantee.

Changelog

• 25cda94 chore(release): prepare v0.1.5 notes
• d70697a fix: raise cosine dedup threshold to prevent same-domain false UPDATE
• da43bfa fix: remove 'not' from negation words, gate conflict check at similarity>=0.7
• 439ca22 fix: sort diff matches by similarity
• 8dc68a9 fix: use Jaccard similarity in dedup to prevent same-domain false UPDATE