toolnet-memory 0.3.15 → 0.3.17

package/.env.example

@@ -59,3 +59,77 @@ TOOLNET_API_TOKEN=
 # ToolNet Memory Hub
 TOOLNET_HUB_OWNER=owner
 TOOLNET_HUB_MAX_EVENTS=100
+
+# ------------------------------------------------------------
+# Graph UI
+# ------------------------------------------------------------
+#
+# Localhost-only by default.
+TOOLNET_GRAPH_HOST=127.0.0.1
+TOOLNET_GRAPH_PORT=9749
+# Optional API bearer token.
+# Recommended whenever Graph UI is reachable from another host.
+# Generate a strong random token, for example:
+#   openssl rand -hex 32
+#
+# TOOLNET_GRAPH_TOKEN=
+# Optional comma-separated Host allow-list for reverse proxies
+# or wildcard binds.
+#
+# TOOLNET_GRAPH_ALLOWED_HOSTS=graph.example.com
+
+# ------------------------------------------------------------
+# Optional remote client-side encryption
+# ------------------------------------------------------------
+#
+# OFF by default. Normal ToolNet usage needs no encryption key.
+#
+# Enable only when you want ToolNet to encrypt objects before
+# they are sent to S3 / R2 / Hugging Face object storage.
+#
+# TOOLNET_REMOTE_ENCRYPTION=on
+#
+# Generate a 32-byte key:
+#   openssl rand -base64 32
+#
+# Configure ONE:
+#
+# TOOLNET_REMOTE_ENCRYPTION_KEY=
+#
+# or preferably from a protected file / Docker secret:
+#
+# TOOLNET_REMOTE_ENCRYPTION_KEY_FILE=/run/secrets/toolnet_remote_encryption_key
+
+# ------------------------------------------------------------
+# Automatic garbage collection
+# ------------------------------------------------------------
+#
+# Disabled by default.
+#
+# When enabled on the ToolNet service, GC runs only for already
+# initialized projects observed by the service.
+#
+# TOOLNET_AUTO_GC=on
+#
+# Default: 168 hours = weekly.
+#
+# TOOLNET_AUTO_GC_INTERVAL_HOURS=168
+#
+# Internal due-check cadence. Default: 15 minutes.
+#
+# TOOLNET_AUTO_GC_TICK_MINUTES=15
+#
+# Remote snapshot deletion remains OFF unless explicitly enabled.
+#
+# TOOLNET_AUTO_GC_REMOTE=off
+#
+# Retention policy overrides:
+#
+# TOOLNET_AUTO_GC_KEEP_SNAPSHOTS=10
+# TOOLNET_AUTO_GC_RUNTIME_DAYS=30
+# TOOLNET_AUTO_GC_STALE_LOCK_MINUTES=10
+# TOOLNET_AUTO_GC_SNAPSHOT_DAYS=30
+#
+# Optional audit actor identity:
+#
+# TOOLNET_AGENT_ID=

package/.release-target

@@ -1 +1 @@
-0.3.15
+0.3.17

package/README.md

@@ -11,7 +11,7 @@
 
 **One project. Multiple coding agents. Continuous context.**
 
-Current release: **v0.3.15**
+Current release: v0.3.17
 
 </div>
 
@@ -28,7 +28,7 @@ ToolNet Memory combines four layers:
 - **Fast project context** — small local startup context for normal work.
 - **Work continuity** — current goal, phase, blockers, decisions, and next actions.
 - **Durable memory** — filtered project knowledge that survives agent/session changes.
-- **Code intelligence** — symbols, dependencies, call graphs, architecture, semantic search, impact analysis, and graph visualization.
+- **Code intelligence** — symbols, dependencies, call graphs, architecture, local SQLite FTS5/BM25 code search, impact analysis, and graph visualization.
 
 ToolNet Memory is **not a raw transcript dump**. Raw session history is kept separate from durable project memory and is protected from normal agent reads.
 
@@ -154,6 +154,25 @@ Project identity is stored in:
 .toolnet/project.json
 ```
 
+For Git projects, new project identity is derived from normalized Git
+repository identity instead of absolute checkout path.
+
+A fresh clone can recover a registered existing ToolNet project identity from
+the configured remote identity registry.
+
+Pre-registry legacy projects require explicit adoption when identity cannot be
+proven automatically:
+
+```text
+toolnet-memory init --adopt-remote <remote-name>
+```
+
+To intentionally initialize without remote identity lookup:
+
+```text
+toolnet-memory init --no-remote-identity
+```
+
 ### 4. Build project intelligence
 
 ```bash
@@ -171,7 +190,7 @@ Type Resolution
     ↓
 Rich Graph
     ↓
-Semantic Code Index
+Local Code Search — SQLite FTS5/BM25
     ↓
 Architecture Intelligence
     ↓
@@ -595,6 +614,161 @@ toolnet-memory project:manual-sync
 
 Secrets and credentials should remain in `.env` or another secret store, not in `PROJECT.md`.
 
+### Optional remote encryption
+
+Remote S3 / R2 / Hugging Face objects can be encrypted client-side with
+AES-256-GCM. Encryption remains disabled by default:
+
+```bash
+export TOOLNET_REMOTE_ENCRYPTION=on
+export TOOLNET_REMOTE_ENCRYPTION_KEY="$(openssl rand -base64 32)"
+```
+
+Existing plaintext remote objects remain readable. New and rewritten
+remote objects are encrypted while the feature is enabled. For services or
+containers, `TOOLNET_REMOTE_ENCRYPTION_KEY_FILE` can load the key from a
+protected file instead of placing the key directly in an environment
+variable. See [`docs/remote-encryption.md`](docs/remote-encryption.md).
+
+---
+
+### Graph UI security
+
+`toolnet-memory graph` remains localhost-first:
+
+```text
+127.0.0.1:9749
+```
+
+For remote exposure, Graph API endpoints support an optional bearer token:
+
+```bash
+export TOOLNET_GRAPH_TOKEN="$(openssl rand -hex 32)"
+export TOOLNET_GRAPH_HOST="0.0.0.0"
+toolnet-memory graph
+```
+
+The browser asks for the token when a protected Graph API returns HTTP 401 and
+keeps it in sessionStorage for that browser session.
+
+`/api/health` stays unauthenticated but returns only a minimal service status.
+
+`TOOLNET_GRAPH_TOKEN` controls Graph API access only. It is not an encryption
+key and is not required for normal localhost usage.
+
+
+## Runtime Capability Truth
+
+ToolNet Memory is local-first and deterministic.
+
+| Component            | Required | Notes |
+| -------------------- | -------- | ----- |
+| LLM                  | no       |       |
+| Embedding provider   | no       |       |
+| Vector database      | no       |       |
+| Mandatory crypto key | no       |       |
+
+Automatic memory promotion and conflict handling are deterministic and evidence-based.
+
+### Local Code Search
+
+The current code-search implementation is:
+
+```text
+Local Code Search — SQLite FTS5/BM25
+Index:   SQLite FTS5
+Ranking: BM25
+Mode:    lexical
+```
+
+It does not use embeddings or a vector database.
+
+The following historical names remain compatibility aliases and are not being
+removed in this release:
+
+- `SemanticCodeEngine`
+- `semantic_code_search`
+- `toolnet-memory semantic`
+- `semantic-index`
+
+### Parser support
+
+| Language family              | Status      |
+| ---------------------------- | ----------- |
+| TypeScript / TSX             | supported   |
+| JavaScript / JSX / MJS / CJS | supported   |
+| Python                       | unsupported |
+| Go                           | unsupported |
+| Rust                         | unsupported |
+| C / C++                      | unsupported |
+
+The supported TypeScript/JavaScript family uses the TypeScript Compiler API.
+
+Tree-sitter is not implemented in this release.
+
+### Cross-machine project identity
+
+Existing `.toolnet/project.json` identity remains canonical.
+
+New Git projects use normalized Git repository identity instead of absolute
+checkout path. A registered existing ToolNet project can therefore be adopted
+by another checkout of the same Git repository.
+
+For a pre-registry legacy project, adoption is explicit:
+
+```text
+toolnet-memory init --adopt-remote <remote-name>
+```
+
+ToolNet refuses unverified silent adoption when repository identity cannot be
+proven.
+
+### Multi-host behavior
+
+Cross-host convergence is implemented under `src/multi-host/**` using immutable
+append-only operations plus deterministic reduction.
+
+The removed legacy `src/sync/**` scaffold was not the active sync engine.
+
+ToolNet does not claim a WebSocket real-time sync protocol or vector-clock CRDT
+implementation.
+
+### Storage truth
+
+**Supported:**
+
+- Local
+- Hugging Face S3-compatible storage
+- S3-compatible storage
+- Cloudflare R2 through the S3-compatible provider
+
+**Unsupported:**
+
+- Google Drive storage
+- GitHub storage backend
+
+### Security truth
+
+ToolNet provides secret scanning and durable-data sanitization.
+
+ToolNet does not provide mandatory client-side encryption. Remote
+client-side encryption is optional and disabled by default. Normal ToolNet
+usage does not require an encryption key. When
+`TOOLNET_REMOTE_ENCRYPTION=on` is configured together with
+`TOOLNET_REMOTE_ENCRYPTION_KEY` (or the `_FILE` variant), ToolNet encrypts
+new remote object writes with AES-256-GCM before they reach S3 / R2 /
+Hugging Face storage. Existing plaintext remote objects remain readable.
+
+See [`docs/remote-encryption.md`](docs/remote-encryption.md).
+
+Project instruction files are treated as untrusted project data, not as system
+or developer authority.
+
+For the detailed capability matrix, see:
+
+- `docs/runtime-truth.md`
+- `docs/repository-capabilities.md`
+
 ---
 
 ## Code Intelligence
@@ -611,11 +785,33 @@ Capabilities include:
 - subsystem clustering,
 - hotspots,
 - dead-code candidates,
-- semantic code search,
+- local lexical code search (SQLite FTS5/BM25),
 - dependency paths,
 - change-impact analysis,
 - visualization datasets.
 
+### Local Code Search
+
+ToolNet code search is deterministic and local.
+
+```text
+Index backend: SQLite FTS5
+Ranking:       BM25
+Mode:          lexical
+Embeddings:    no
+Vector DB:     no
+LLM/model:     no
+Network:       not required
+```
+
+The historical public names remain available for compatibility:
+
+- CLI: `toolnet-memory semantic`
+- MCP: `semantic_code_search`
+- API: `SemanticCodeEngine`
+
+Those names are compatibility aliases and do not indicate embedding-based semantic retrieval.
+
 Useful commands:
 
 ```bash
@@ -625,7 +821,7 @@ toolnet-memory index
 # Incremental update
 toolnet-memory incremental
 
-# Semantic search
+# Local lexical code search (legacy command name)
 toolnet-memory semantic "authentication flow"
 
 # Change impact
@@ -939,7 +1135,7 @@ ToolNet Memory uses GitHub Actions for CI and npm releases.
 The current release line is:
 
 ```text
-npm package: toolnet-memory@0.3.14
+npm package: toolnet-memory@0.3.17
 Git tag:     v0.3.14
 ```
 
@@ -965,3 +1161,117 @@ See [CHANGELOG.md](CHANGELOG.md) for release history.
 ## License
 
 MIT © 2026 LBT-AI. See [LICENSE](LICENSE).
+
+
+
+Docker
+
+ToolNet Memory includes a hardened Node.js 22 container build.
+
+docker build -t toolnet-memory:local .
+docker run --rm toolnet-memory:local --version
+
+Run the daemon with persistent ToolNet runtime state:
+
+docker run --rm \
+  -v toolnet-memory-home:/home/node/.toolnet \
+  toolnet-memory:local
+
+Run against the current project:
+
+docker run --rm \
+  -v "$PWD:/workspace" \
+  -v toolnet-memory-home:/home/node/.toolnet \
+  toolnet-memory:local \
+  doctor
+
+The final image runs as a non-root user and the Docker health check uses the
+ToolNet daemon's real Unix-socket ping protocol.
+
+Release image target:
+
+ghcr.io/lbt-ai/toolnet-memory
+
+Optional Docker Hub publishing supports:
+
+lbtai/toolnet-memory
+
+when Docker Hub publishing is explicitly enabled in repository settings.
+
+See docs/docker.md for Compose, shared-volume dedupe, permissions, health
+checks, and multi-architecture release details.
+
+
+Standalone binaries
+
+ToolNet Memory can run without a separately installed Node.js/npm runtime.
+
+Release targets:
+
+Linux   x64 / arm64
+macOS   x64 / arm64
+Windows x64
+
+Example:
+
+```bash
+chmod +x toolnet-memory-linux-x64
+./toolnet-memory-linux-x64 --version
+```
+
+Standalone binaries embed the Node.js runtime and ToolNet production
+dependency graph. The npm package remains fully supported.
+
+See [`docs/standalone.md`](docs/standalone.md) for platform support,
+service limitations, checksums, and build instructions.
+
+
+
+Audit log and automatic GC
+
+ToolNet can maintain a tamper-evident project audit history:
+
+.toolnet/audit/events.jsonl
+
+Inspect and verify:
+
+```bash
+toolnet-memory audit
+toolnet-memory audit:verify
+```
+
+The log records durable memory writes, snapshot restore/recovery, Guard
+checks, and GC execution without intentionally storing raw memory
+content or raw shell commands.
+
+Automatic GC remains opt-in:
+
+```bash
+export TOOLNET_AUTO_GC=on
+toolnet-memory service
+```
+
+The default interval is weekly. Remote snapshot GC remains disabled
+unless `TOOLNET_AUTO_GC_REMOTE=on` is explicitly configured.
+
+See [`docs/audit-log.md`](docs/audit-log.md) and
+[`docs/auto-gc.md`](docs/auto-gc.md).
+
+
+
+Non-TypeScript local code search
+
+Structural graph parsing remains TypeScript/JavaScript-only.
+
+Python, Go, Rust, C, and C++ files are now included in deterministic local
+code search through sanitized file chunks indexed by SQLite FTS5/BM25.
+
+```bash
+toolnet-memory code:capabilities
+```
+
+Optional external LSP installations are detected for capability reporting
+only. ToolNet does not auto-download or automatically execute language
+servers during indexing.
+
+See [`docs/non-ts-intelligence.md`](docs/non-ts-intelligence.md).

package/bin/toolnet-memory

@@ -80,6 +80,18 @@ case "$COMMAND" in
     shift
     run bundle/gc.js src/retention/cli.ts "$@"
     ;;
+  audit)
+    shift
+    run bundle/audit.js src/audit/cli.ts "$@"
+    ;;
+  audit:verify)
+    shift
+    run bundle/audit.js src/audit/cli.ts verify "$@"
+    ;;
+  code:capabilities)
+    shift
+    run bundle/code-capabilities.js src/code-intelligence/parsers/capability-cli.ts "$@"
+    ;;
 
   run)
     run bundle/runtime.js src/index.ts