SASHI - Smart AI Shell Interface

Local-first AI assistant powered by Ollama/Llama. Privacy-first, no data leaves your machine. Runs on modest hardware (i7, 8GB RAM, no GPU). Optional cloud fallback via OpenRouter.

---

Table of Contents

• Overview
• Performance Tuning
• Architecture
• Models
• MCP Structure
• Installation
• Usage
• Termux (Mobile)
• SQLite Schema
• Aliases Reference
• Smart Push
• Session Timeline

---

Overview

SASHI routes all queries through ollama run (native CLI, streaming, model stays hot). No cloud dependency. No API keys required for local use.

┌─────────────────────────────────────────────────────────────┐
│                      USER INPUT                              │
│            text / voice / pipe / interactive                 │
└─────────────────────────────────────────────────────────────┘
                            │
                            ▼
┌─────────────────────────────────────────────────────────────┐
│                     SASHI v3.2.2                             │
│  ┌─────────┐  ┌─────────┐  ┌─────────┐  ┌─────────┐        │
│  │ Router  │→ │ Logger  │→ │ History │→ │ Output  │        │
│  └─────────┘  └─────────┘  └─────────┘  └─────────┘        │
└─────────────────────────────────────────────────────────────┘
        │              │              │              │
        ▼              ▼              ▼              ▼
┌───────────┐  ┌───────────┐  ┌───────────┐  ┌───────────┐
│  Llama    │  │OpenRouter │  │   Gmail   │  │   Voice   │
│  Ollama   │  │  (Cloud)  │  │    API    │  │  Google   │
│  (Local)  │  │ (Fallback)│  │  (OAuth)  │  │   STT     │
└───────────┘  └───────────┘  └───────────┘  └───────────┘
   Primary       Optional        Context        Input

Process Map

Data Flow

Smart Push Flow

---

Performance Tuning

Benchmarked and optimized for CPU-only hardware (no GPU). These settings were proven across 3 rounds of benchmarking on an i7-6500U (2 cores, 4 threads, 7.6GB RAM, 8GB swap).

Why These Settings Matter

Setting	Wrong Value	Right Value	What Happens
num_thread	4 (all threads)	2 (physical cores)	HT contention = 30% slower with 4 threads
CPU governor	powersave	performance	Prevents CPU throttling mid-inference
OLLAMA_MAX_LOADED_MODELS	default	1	Prevents 2 models fighting for RAM
OLLAMA_KEEP_ALIVE	5m	30m	Model stays hot between queries
OLLAMA_NUM_PARALLEL	default	1	Single user = no parallel overhead

Apply Optimizations (one time)

# 1. CPU governor (needs sudo)
echo performance | sudo tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor

# 2. Ollama service tuning (needs sudo)
sudo mkdir -p /etc/systemd/system/ollama.service.d
cat << 'OVR' | sudo tee /etc/systemd/system/ollama.service.d/override.conf
[Service]
Environment="OLLAMA_NUM_PARALLEL=1"
Environment="OLLAMA_MAX_LOADED_MODELS=1"
Environment="OLLAMA_KEEP_ALIVE=30m"
OVR
sudo systemctl daemon-reload && sudo systemctl restart ollama

# 3. Build optimized models (no sudo)
ollama create fast-sashi -f ~/ollama-local/Modelfile.fast
ollama create sashi-llama -f ~/ollama-local/Modelfile.system

Benchmark Proof

Tested with identical prompts (coding, reasoning, knowledge). Eval rate = tokens/second during generation.

Why num_thread 2 beats num_thread 4:

num_thread	3B tok/s	8B tok/s	Notes
4 (all threads)	2.8	3.0	HT contention kills throughput
2 (physical cores)	4.0	3.7	+43% on 3B, +23% on 8B

On a 2-core/4-thread CPU, hyperthreading competes for the same execution units. LLM inference is pure compute — it needs real cores, not virtual ones.

Hardware Speed Ceiling

Model	Eval Rate	Cold Start	Hot Query Time
sashi-llama (3B)	~4.0 tok/s	5s	~35s
sashi-llama-8b (8B)	~3.7 tok/s	60s (swap)	~40s

The bottleneck is memory bandwidth, not CPU clock. This is the ceiling for this hardware.

---

Architecture

BDPM Governance Layer (v3.2.2)

The 4-layer BDPM governance model spans both repos. See the full swimlane diagram in kanban-pmo/docs/diagrams/bdpm-swimlanes.svg:

• Business — kanban-pmo intake, sprint planning, milestone gates
• Development — git push, model build, test verify, smart-push (this repo)
• Production — gRPC pipeline dispatch, ollama inference, file write, DB log (this repo)
• Monitoring — cred audit, health check, doc sync, CMMI compliance

Credential Layer

persist-memory-probe/lib/sh/gatekeeper_3_1_0.sh is the credential gateway. It delegates inference to ollama-local while owning github/sign/remote routes. All sashi sub-commands (kanban, probe, write) pass through the gatekeeper before hitting the gRPC layer.

Directory Structure

ollama-local/
├── sashi                    # Main CLI (v3.2.2)
├── .env                     # Config (LOCAL_MODEL, OLLAMA_HOST)
├── .env.termux              # Termux override (llama3.2:1b)
├── Modelfile.fast            # 3B fast model (concise system prompt, default)
├── Modelfile.system         # 3B full model config (comprehensive system prompt)
├── Modelfile.8b             # 8B model config (num_thread 2, system prompt) [archived]
├── install.sh               # One-command installer
├── docker-compose.yml       # Container orchestration
│
├── db/
│   └── history.db           # SQLite WAL (10 tables, 27 indexes)
│
├── docs/
│   ├── termux-ollama-plan.md  # Model sizing, swap analysis
│   ├── termux-setup.md        # Phone install guide
│   └── monetization.md        # Revenue playbook
│
├── mcp/                     # Model Context Protocol
│   ├── claude/              # Claude Code integration
│   ├── llama/               # Local llama tools, ai-orchestrator
│   ├── gmail/               # Gmail CLI (search, recent, export)
│   └── voice/               # Voice input (CLI + GUI)
│
├── lib/
│   └── sh/
│       ├── banner.sh        # sashi_banner() ASCII art — sourced by all tools
│       ├── aliases.sh       # Shell aliases — 80+ total incl. 30 filesystem (v3.2.2)
│       ├── usb-monitor.sh   # USB vendor DB + sysfs scanner
│       └── wifi-debug.sh    # ADB WiFi library
│
├── scripts/
│   ├── smart-push.sh        # 424-line git automation
│   ├── rebuild-models.sh    # Rebuild fast-sashi + sashi-llama-8b from Modelfiles
│   ├── android-setup.sh     # Downloads + installs Android SDK, platform-tools, adb
│   ├── termux-sync.sh       # Desktop ↔ phone sync
│   ├── git-setup.sh         # SSH/GitHub setup
│   └── git-aliases.sh       # Git alias installer
│
└── old-archive/             # Archived sessions (never deleted)

Models

Available Models

Model	Params	Size	Speed	Modelfile	Use Case
fast-sashi	3B	2.0GB	4.0 tok/s	Modelfile.fast	Default — concise, date-aware
sashi-llama	3B	2.0GB	4.0 tok/s	Modelfile.system	Full system prompt, verbose context
sashi-llama-8b	8B	4.9GB	3.7 tok/s	Modelfile.8b	Better quality, needs swap
llama3.2:1b	1B	1.3GB	fast	(base)	Termux/mobile — lightweight

Model Selection

# Desktop default (fast-sashi, 3B, concise)
sashi ask "explain TCP"

# Full context model (bigger system prompt)
ollama run sashi-llama "explain TCP in detail"

# Switch to 8B for quality (edit .env: LOCAL_MODEL=sashi-llama-8b)
ollama run sashi-llama-8b "explain TCP in detail"

# Termux auto-detects and uses 1B
# (handled by .env.termux override)

All custom models include system prompts with date awareness. The default fast-sashi is concise; sashi-llama has full hardware/file/alias context.

---

MCP Structure

MCP Groups

Category	Name	Type	Description
Core	sashi	CLI	Main router and interface
Model	llama	Local	Llama 3.2/3.1 via Ollama — primary
Model	claude	Integration	Claude Code CLI — complex tasks
Automation	pipeline	gRPC	Inference + file write + orchestration
Protocol	voice	Input	Google Speech-to-Text
Protocol	gmail	Context	Gmail API for email data

Route Comparison

Route	Type	Speed	Context	Cost	Use Case
sashi ask	Local (3B)	~4 tok/s	4K	Free	Quick queries
sashi ask (8B)	Local (8B)	~3.7 tok/s	4K	Free	Better reasoning
sashi online	Cloud (OpenRouter)	~2s	varies	Free tier	When local isn't enough

---

Installation

One-liner (recommended)

curl -fsSL https://raw.githubusercontent.com/tmdev012/ollama-local/main/install.sh | bash

Then reload your shell:

source ~/.bashrc   # or source ~/.zshrc
sashi status

The installer handles everything: Ollama, llama3.2 model, CPU tuning, DB init, shell aliases.

Flags

# Skip model download (if you already have llama3.2)
curl -fsSL .../install.sh | bash -s -- --no-models

# Skip CPU governor tuning (e.g. on a server)
curl -fsSL .../install.sh | bash -s -- --no-gpu-tune

# Termux / Android
curl -fsSL .../install.sh | bash -s -- --termux

Manual Install (Linux)

# 1. Install Ollama
curl -fsSL https://ollama.ai/install.sh | sh
sudo systemctl enable --now ollama

# 2. Clone repo
git clone https://github.com/tmdev012/ollama-local.git ~/ollama-local

# 3. Pull base model + build optimized custom models
ollama pull llama3.2
ollama create fast-sashi -f ~/ollama-local/Modelfile.fast

# 4. (Optional) 8B model — needs 8GB+ swap
ollama pull llama3.1:8b
ollama create sashi-llama-8b -f ~/ollama-local/Modelfile.8b

# 5. Apply performance tuning (see Performance Tuning section above)

# 6. Add to shell
echo 'export PATH="$HOME/ollama-local:$PATH"' >> ~/.bashrc
echo 'source ~/ollama-local/lib/sh/aliases.sh' >> ~/.bashrc
source ~/.bashrc

Docker

docker-compose up -d
docker exec -it sashi-ai bash

---

Usage

Basic Commands

# Quick question (local llama 3B)
sashi ask "What is Python?"
sask "Explain REST APIs"

# 8B quality route (better reasoning)
sashi 8b "Explain async/await in depth"
s8b "complex question"

# Code help
sashi code "Write a sorting function in Python"
scode "Write a sorting function in Python"

# Interactive chat (streams via ollama run)
sashi chat
schat

# Cloud fallback (when local isn't enough)
sashi online "Explain quantum computing in depth"
sonline "complex question here"

# Voice input
sashi voice              # Single prompt
sashi voice --continuous # Keep listening
sashi voice --gui        # Desktop app

# System status
sashi status    # or: sstatus — shows gRPC health + latest changelog
sashi models    # or: smodels
sashi history   # or: shistory
sashi changelog # View CHANGELOG.md inline

gRPC Daemon Management

# Start both gRPC servers (:50051 kanban-pmo, :50052 probe)
sashi grpc start

# Check daemon health
sashi grpc status

# Stop / restart
sashi grpc stop
sashi grpc restart

# Tail logs
sashi grpc logs

Probe CLI (via :50052)

# Sync a repo into probe.db
sashi probe sync [repo]

# List registered repos
sashi probe list

# Get credential recommendation
sashi probe recommend <operation>

# Export training dialogs
sashi probe export [N]

# Write a file via gRPC
sashi probe write <path> <content>

# Check probe server health
sashi probe status

IDE

# Launch terminal Android/Kotlin IDE
sashi ide [project-path]

Kanban

sashi kanban board    # Full board view
sashi kanban wip      # In-progress cards
sashi kanban backlog  # Backlog
sashi kanban state    # Summary counts

Pipe Support

cat code.py | sashi code "explain this"
git diff | sashi code "review this"
cat README.md | summarize

Git Pipeline

smartpush              # Full interactive smart commit (sp, gpush)
gitpush "message"      # Add + Commit + Push (gpp, ship)
ghist                  # View commit history from SQLite
gver                   # List version tags
gissue 42              # Find commits by issue number

---

Version History

Version	Date	Key Change
v1.0	2026-02	Initial CLI, ollama run calls
v2.0	2026-02-05	HTTP API optimization, 5-8s→2.2s, voice, 22 clean aliases
v3.0	2026-02-08	Back to ollama run (streams, keeps model hot), DeepSeek removed
v3.1	2026-02-19	banner.sh, aliases.sh, kanban subcommand, smart-push, gRPC stubs
v3.2.0	2026-02-22	gRPC daemon manager, probe CLI, IDE, 8B routing, 245 training dialogs
v3.2.1	2026-03-01	sashi usb/wifi/hf, USB vendor DB, WiFi ADB, HuggingFace fallback
v3.2.2	2026-03-01	30 advanced filesystem aliases, sashi wallog (Modelfile ↔ WAL log)

v3.2.1→v3.2.2 Changes

Aspect	v3.2.1	v3.2.2
Filesystem aliases	none	30 aliases across 9 categories (find/disk/list/archive/copy/perms/symlink/checksum/watch)
WAL log command	none	sashi wallog [N] — Modelfile git log + SQL changelog + commits + WAL checkpoint
Alias shortcut	—	swallog
SVGs	v3.2.0 labels	v3.2.2 across bdpm-swimlanes, gazette-architecture, process-map
Router label	v2.0	v3.2.2 (process-map-animated.svg)

v3.1→v3.2 Changes

Aspect	v3.1	v3.2
gRPC	hollow ProbeSyncServicer stub	daemon manager :50051 + :50052 via sashi grpc
Probe CLI	sync-only via kanban-pmo	full probe CLI via :50052 (list/recommend/export/write)
IDE	none	sashi ide — terminal Android/Kotlin IDE (rich TUI, ADB)
8B routing	manual model switch	sashi 8b <prompt> — direct 8B quality route
Training data	0 dialogs	245 dialogs in probe.db (multi_ternary, filewrite_grpc, android_ide)
Modelfiles	v4.0	v4.1 — gRPC + probe + IDE + Android docs in system prompt

---

Termux (Mobile)

Sashi auto-detects Termux and switches to a lighter model.

Quick Setup

# In Termux (install from F-Droid, NOT Play Store)
pkg update && pkg install git openssh
git clone git@github.com:tmdev012/ollama-local.git ~/ollama-local

# Option A: Local ollama on phone (6GB+ RAM)
pkg install golang cmake git
# Build ollama from source, then:
ollama pull llama3.2:1b
ln -sf ~/ollama-local/sashi ~/bin/sashi
sashi ask "hello from my phone"

# Option B: Cloud route (any phone)
# Set OPENROUTER_API_KEY in .env
sashi online "hello from my phone"

How Auto-Detection Works

# In sashi (line 8):
[[ -n "$TERMUX_VERSION" || -d "/data/data/com.termux" ]] && source .env.termux

# .env.termux sets:
LOCAL_MODEL=llama3.2:1b    # 1B instead of 3B

Model Sizing for Phones

Phone RAM	Model	Size	Speed
3-4GB	smollm2:1.7b	1.0GB	Fast
4-6GB	llama3.2:1b	1.3GB	Fast
6-8GB	llama3.2:3b	2.0GB	Medium

See docs/termux-ollama-plan.md for full analysis.

---

SQLite Schema

ERD Diagram

┌─────────────────────────────────────┐
│            queries                  │
├─────────────────────────────────────┤
│ PK id              INTEGER          │
│    timestamp       DATETIME         │
│    model           TEXT        ◄────┼─── idx_queries_model
│    prompt          TEXT             │
│    response_length INTEGER          │
│    duration_ms     INTEGER     ◄────┼─── idx_queries_duration
│                               ◄─────┼─── idx_queries_timestamp
└───────────────┬─────────────────────┘
                │ 1:N
┌───────────────▼─────────────────────┐
│           favorites                 │
├─────────────────────────────────────┤
│ PK id              INTEGER          │
│ FK query_id        INTEGER     ◄────┼─── idx_favorites_query
│    label           TEXT             │
└─────────────────────────────────────┘

┌─────────────────────────────────────┐
│          mcp_groups                 │
├─────────────────────────────────────┤
│ PK id              INTEGER          │
│    name            TEXT (UNIQUE)    │
│    category        TEXT        ◄────┼─── idx_mcp_groups_category
│    description     TEXT             │
│    config_path     TEXT             │
│    enabled         INTEGER     ◄────┼─── idx_mcp_groups_enabled
│    created_at      DATETIME         │
│    updated_at      DATETIME         │
└─────────────────────────────────────┘

┌─────────────────────────────────────┐
│            commits                  │
├─────────────────────────────────────┤
│ PK id              INTEGER          │
│    hash            TEXT        ◄────┼─── idx_commits_hash
│    message         TEXT             │
│    auto_description TEXT            │
│    issue_number    TEXT        ◄────┼─── idx_commits_issue
│    version_tag     TEXT        ◄────┼─── idx_commits_version
│    branch          TEXT             │
│    files_changed   INTEGER          │
│    lines_added     INTEGER          │
│    lines_deleted   INTEGER          │
│    categories      TEXT             │
│    timestamp       DATETIME         │
│    tree_backup     TEXT             │
└─────────────────────────────────────┘

Tables

Table	Rows	Indexes	Purpose
queries	N	3	AI query history
favorites	N	1	Starred queries
mcp_groups	6	2	MCP provider registry
commits	N	5	Smart push commit tracking

Indexes (11 total)

-- queries
CREATE INDEX idx_queries_model ON queries(model);
CREATE INDEX idx_queries_timestamp ON queries(timestamp);
CREATE INDEX idx_queries_duration ON queries(duration_ms);

-- favorites
CREATE INDEX idx_favorites_query ON favorites(query_id);

-- mcp_groups
CREATE INDEX idx_mcp_groups_category ON mcp_groups(category);
CREATE INDEX idx_mcp_groups_enabled ON mcp_groups(enabled);

-- commits
CREATE INDEX idx_commits_hash ON commits(hash);
CREATE INDEX idx_commits_version ON commits(version_tag);
CREATE INDEX idx_commits_issue ON commits(issue_number);
CREATE INDEX idx_commits_branch ON commits(branch);
CREATE INDEX idx_commits_timestamp ON commits(timestamp);

---

Aliases Reference

SASHI (AI)

Alias	Command	Description
s	sashi	Main interface
sask	sashi ask	Quick question (local 3B)
s8b	sashi 8b	8B quality route
shf	sashi hf	HuggingFace Inference API (free tier)
scode	sashi code	Code help (local llama)
slocal	sashi local	Same as ask
schat	sashi chat	Interactive chat
sstatus	sashi status	System status + gRPC health
smodels	sashi models	List models
shistory	sashi history	Query history
schangelog	sashi changelog	Full CHANGELOG.md
swallog	sashi wallog [N]	Modelfile git log + SQL WAL changelog
skanban	sashi kanban board	Kanban board
sgmail	sashi gmail	Email context
usb-scan	sashi usb scan	List USB devices with vendor names
usb-watch	sashi usb watch	Real-time USB plug/unplug events
wifi-init	sashi wifi init	ADB WiFi: tcpip + auto IP detect
wifi-status	sashi wifi status	List wireless ADB devices

Filesystem (v3.2.2 — 30 aliases)

Find & Filter

Alias	Expands to	Example
ff	find . -type f -name	ff "*.log"
ffd	find . -type d -name	ffd "build*"
ffl	find . -type l	list all symlinks
fmod	find . -type f -mmin	fmod -60 (last 60 min)
fsize	find . -type f -size	fsize +100M
fnew	find . -type f -newer	fnew ref-file
fdup	md5sum dedup pipeline	find duplicate files
fempty	find . -type f -empty	zero-byte files
fdangling	find broken symlinks	dangling symlink scan

Disk Analysis

Alias	Expands to	Notes
duh	du -sh * | sort -rh	current dir, size-sorted
dua	du -ah --max-depth=1 | sort -rh	all entries, depth 1
dut	du -sh */	dirs only, sorted
dfh	df -hT --exclude-type=tmpfs …	real disks only
dfio	df -i	inode usage

Directory Listing

Alias	Expands to	Notes
lsl	ls -lahF --color	full listing
lst	ls -lath --color	sorted by time
lsz	ls -laSh --color	sorted by size
lsd	ls -lah --group-directories-first	dirs first
lsr	ls -lahR --color	recursive

Archive

Alias	Usage	Notes
tarc	tarc out.tar.gz dir/	create tar.gz
tarx	tarx file.tar.gz	extract
tarxv	tarxv file.tar.gz	extract verbose
tarl	tarl file.tar.gz	list contents
tarbz	tarbz out.tar.bz2 dir/	bzip2
zipr	zipr out.zip dir/	zip recursive

Copy / Move / Delete

Alias	Expands to	Notes
cpv	rsync -ah --progress	cp with progress bar
cpvr	rsync -ahr --progress --delete	mirror dir
mvv	mv -v	verbose move
rmv	rm -iv	interactive + verbose
rmrf	rm -rf	explicit destructive intent

Permissions & Ownership

Alias	Expands to	Notes
chmodr	chmod -R	recursive chmod
chownr	chown -R	recursive chown
mkexec	chmod +x	make executable
fixperms	find -exec chmod 644/755	fix file/dir perms

Symlinks

Alias	Expands to	Notes
lnr	ln -sr	relative symlink
lna	ln -sf	absolute/force symlink
lslinks	find -type l -exec ls	show all symlinks + targets

Checksum & Compare

Alias	Expands to	Notes
fhash	sha256sum	hash a file
fcheck	sha256sum -c	verify checksum file
mdiff	diff -rq	compare two directories
mdiffu	diff -ru	unified diff of dirs

Watch & Monitor

Alias	Expands to	Notes
fwatch	watch -n1 "ls -lah"	watch dir every 1s
fwatchp	inotifywait -rm -e modify,create,delete,move	inotify on path

Git

Alias	Command	Description
gs	git status -sb	Short status
gd	git diff	Show diff
gl	git log --oneline -20	Short log
ga	git add	Stage files
gaa	git add -A	Stage all
gc	git commit -m	Commit
gp	git push	Push
gpl	git pull	Pull
gb	git branch	Branches
gco	git checkout	Checkout

Git Pipeline

Alias	Description
gitpush "msg"	Add + Commit + Push
gpp "msg"	Short for gitpush
ship "msg"	Another alias
gship	Interactive (prompts for message)

Smart Push (v2.0)

Alias	Description
smartpush	Full interactive smart commit
sp	Short alias for smartpush
gpush	Another alias
ghist	View commit history from SQLite
gver	List all version tags
gissue "N"	Find commits by issue number

Ollama

Alias	Command	Description
ollama-up	systemctl start ollama	Start service
ollama-down	systemctl stop ollama	Stop service
ollama-status	Check status + list	Status
ollama-logs	journalctl -u ollama	View logs

Pipe Helpers

Alias	Description
analyze	cat file | analyze
summarize	cat file | summarize
explain	cat file | explain
review	cat file | review

---

Tech Stack

Component	Technology
Shell	Bash / Zsh
Local AI	Ollama + Llama 3.2 (3B) / Llama 3.1 (8B)
Cloud AI	OpenRouter (free tier, fallback)
Database	SQLite 3 (WAL mode, 10 tables, 27 indexes)
Voice	Google Speech-to-Text
Mobile	Termux (Android)
Container	Docker + Compose
VCS	Git + GitHub
Auth	SSH (ED25519)

Dependencies

# System
curl jq python3 sqlite3

# Ollama (required)
ollama (+ llama3.2 model, optionally llama3.1:8b)

# Voice (optional)
portaudio19-dev python3-pyaudio python3-tk
pip3 install SpeechRecognition

---

Termux Sync

Sync shell configs between devices (Linux ↔ Android/Termux).

Usage

# On Linux - backup to GitHub
termux-sync push

# On Termux - restore from GitHub
git clone git@github.com:tmdev012/ollama-local.git
cd ollama-local
./scripts/termux-sync.sh pull

Commands

Command	Description
termux-sync push	Upload configs to GitHub
termux-sync pull	Download configs from GitHub
termux-sync status	Show sync status
termux-sync auto	Enable auto-sync on exit

Synced Files

• ~/.bashrc
• ~/.zshrc
• ~/.bash_history
• ~/.zsh_history
• ~/.gitconfig
• ~/.ssh/config

---

Environment Variables

# .env file
LOCAL_MODEL=llama3.2           # Default local model (or sashi-llama-8b for 8B)
OLLAMA_HOST=http://localhost:11434
OPENROUTER_API_KEY=            # Optional, for cloud fallback

# Git
GIT_USER=tmdev012
GIT_EMAIL=tmdev012@users.noreply.github.com
GIT_REPO=ollama-local

---

Smart Push

Intelligent git commit system with auto-categorization, version tagging, and issue linking.

Features

• Auto-categorization: Files categorized by extension
• Branch comparison: Shows ahead/behind vs main
• Version tagging: Semantic versioning with auto-increment
• Issue linking: Links commits to GitHub issues
• File tree backup: Snapshots before each commit
• SQLite tracking: All commits stored with metadata

File Categories

Category	Extensions
frontend:styles	html, css, scss, sass, less
frontend:script	js, jsx, ts, tsx, vue, svelte
backend:python	py, pyw
scripts:shell	sh, bash, zsh, fish
config	json, yaml, yml, toml, ini, conf, env
database	sql, db, sqlite
docs	md, txt, rst, doc
devops:docker	Dockerfile, docker-compose*
testing	test*, _test., spec.
mcp:module	mcp/* directory

Usage

# Interactive mode
smartpush

# Output includes:
# [1/8] Branch comparison (feature vs main)
# [2/8] File tree backup
# [3/8] File changes by category
# [4/8] Diff summary (+lines/-lines)
# [5/8] Auto-generated description
# [6/8] Commit details (version tag, issue #)
# [7/8] Commit
# [8/8] Push

Query History

# View commit history
ghist

# List version tags
gver

# Find commits by issue
gissue 42

---

Session Timeline

Git Commit History (10-hour session)

Commit	Tag	Description	Files
faaef58	-	Clean: MCP structure with sashi CLI	16
b57005f	-	Add Gmail module for email context	4
b619c56	-	v2.0.0: SASHI optimization, voice, Git/SSH	17
373647c	-	Add termux-sync for cross-device backup	2
d0445aa	-	Add comprehensive README	1
1ff6995	-	Add smart-push v2.0	1
1904374	v0.0.1	Smart alias for YAML webhooks	1
4c1981b	v0.0.2	Filetree update - structure changes	1
bcef945	-	Timestamped filetree monitoring	1
0ef3279	-	MCP module directories consistency	3

Session Stats

Total commits:     10
Files created:     30+
Files modified:    12
Lines added:       4,500+
Lines deleted:     400+
Tables created:    4
Indexes created:   11
Aliases added:     25+
Duration:          ~10 hours

Key Accomplishments

1. MCP Architecture - 5 modules (claude, llama, voice, gmail, pipeline)
2. SASHI v2.0 - HTTP API optimization (5-8s → 2.2s)
3. Voice Input - CLI + GUI with Google Speech-to-Text
4. Smart Push - Auto-categorization, versioning, SQLite tracking
5. Alias Cleanup - 43 broken → 22 clean MCP-aligned
6. SQLite Schema - 4 tables, 11 indexes
7. Git/SSH Setup - ED25519 keys, GitHub auth
8. Docker Support - Full containerization
9. Termux Sync - Cross-device config backup
10. Documentation - README, CHANGELOG, schema docs

---

Contributing

# Clone
git clone git@github.com:tmdev012/ollama-local.git
cd ollama-local

# Make changes
# ...

# Push
gitpush "Description of changes"

---

License

MIT

---

Credits

• Author: tmdev012
• AI Assistant: Claude Opus 4.6 (Anthropic)
• Models: Meta Llama 3.2 (3B), Meta Llama 3.1 (8B)

---

Built with Claude Code CLI - Feb 2026 | Last updated: 2026-02-22