termstory 0.2.11

Local shell history parsing, session grouping, and visual daily chronicle terminal dashboard

TermStory — Your Personal Developer Memory Engine

Parse your shell history. Recover your past. Understand your work.

TermStory turns your terminal history into a searchable, AI-narrated timeline of your development life. It groups shell commands into sessions, correlates Git commits, resolves project names, and renders everything into a high-density TUI dashboard — with a built-in forensic engine that can recover the real dates of commands you typed before you even knew timestamps were missing.

pip install termstory
termstory ui

---

Table of Contents

1. Core Philosophy
2. Installation & Quick Start
3. Project Layout
4. The Ingestion Pipeline
5. Shell History Parsing
6. The Timestamp Detective (v0.2.9)
7. Project Resolution
8. Git Commit Correlation
9. Database Schema
10. Privacy Sanitizer
11. AI Client
12. TUI Dashboard
13. AI Narrative Design
14. CLI Reference
15. Configuration
16. Testing
17. Troubleshooting

---

1. Core Philosophy

TermStory is not a tracking tool, a productivity auditor, or a corporate analytics dashboard. It is a personal developer memory engine built on three ideas:

• Recognize, don't inspect. The goal is instant recognition — "Ah, that was the day I fought the Docker networking bug" — not a wall of cd and ls entries. Noise commands are filtered automatically.
• Density over decoration. No rounded panels, no double borders, no empty margins. Clean column alignment, tight spacing, information-first.
• Screenshot-friendly. Every view fits in one terminal screen and tells a complete, self-contained story.

---

2. Installation & Quick Start

Requirements: Python 3.9+, a terminal with zsh or bash.

pip install termstory

Enable timestamps (zsh only — one time setup)

TermStory works best when your shell records timestamps. If you haven't done this already:

echo '\nsetopt EXTENDED_HISTORY\nsetopt HIST_STAMPS="yyyy-mm-dd"' >> ~/.zshrc
source ~/.zshrc

Already have old history without timestamps? TermStory's Timestamp Detective (v0.2.9) will forensically recover real dates from your git log, filesystem metadata, and package manager artifacts automatically. You don't lose your past.

First launch

termstory ui          # Interactive TUI dashboard (recommended)
termstory today       # Today's timeline in the terminal
termstory search auth # Search across all history

On first launch, TermStory detects whether you have timestamps enabled. If not, it offers to enable EXTENDED_HISTORY automatically. You can skip this and proceed anyway — the Timestamp Detective will handle your legacy history.

---

3. Project Layout

termstory/
├── setup.py                     # Package metadata
├── pyproject.toml               # Build config
├── README.md                    # This document
├── DATA_PRIVACY.md              # LLM data handling policy
├── termstory/
│   ├── __init__.py              # Version: 0.2.9
│   ├── __main__.py              # python3 -m termstory entry point
│   ├── cli.py                   # Typer CLI — all commands & ingestion entry point
│   ├── tui.py                   # Textual TUI dashboard & all widgets
│   ├── parser.py                # Shell history parsing engine
│   ├── timestamp_detective.py   # v0.2.9 — forensic timestamp recovery engine
│   ├── session.py               # 30-minute session grouping
│   ├── project.py               # VCS root detection & project name resolution
│   ├── git_integration.py       # git log subprocess client & commit cleaner
│   ├── database.py              # SQLite layer (WAL, schema, queries, cache)
│   ├── date_utils.py            # Timezone & timestamp utilities
│   ├── sanitizer.py             # Local credential & PII redaction
│   ├── ai.py                    # Zero-dependency LLM client (urllib only)
│   ├── insights.py              # Focus score & pattern calculations
│   ├── models.py                # Command, Session, Project, Commit dataclasses
│   └── formatter.py            # CLI output layout & Rich styling
└── tests/
    ├── fixtures/
    │   └── sample_history.txt
    ├── test_parser.py
    ├── test_session.py
    ├── test_project.py
    ├── test_git_integration.py
    ├── test_database.py
    ├── test_database_queries.py
    ├── test_sanitizer.py
    ├── test_ai.py
    ├── test_tui.py
    ├── test_formatter_rich.py
    ├── test_insights.py
    ├── test_timestamp_detective.py  # v0.2.9 — 155 tests
    └── test_integration.py

---

4. The Ingestion Pipeline

Every CLI command and every TUI launch runs this pipeline:

~/.zsh_history / ~/.bash_history
         │
         ▼
    parser.py  ──────────────────────────────────────────────────────────┐
    Parse raw history into Command objects.                               │
    Separate timestamped commands from legacy (no-timestamp) commands.   │
         │                                                                │
         ▼  (legacy commands only)                                        │
    timestamp_detective.py                                                │
    Phase A: Replay cd/pushd/popd → virtual CWD per command              │
    Phase B: 5 forensic detectors (git log, file stat, pkg mgr,          │
             docker, lockfiles) → real timestamps + Chain of Custody      │
    Phase C: Anchor Interpolation → linear fill between anchors          │
         │                                                                │
         └─────────────────────────────────────────────────────────────►─┤
         ▼                                                                │
    session.py                                                            │
    Group commands into sessions (30-minute idle threshold)              │
         │                                                                │
         ▼                                                                │
    project.py                                                            │
    Detect VCS root → extract project name from manifests → "Other"      │
         │                                                                │
         ▼                                                                │
    git_integration.py                                                    │
    Run git log for session timeframe → clean commit messages            │
         │                                                                │
         ▼                                                                │
    sanitizer.py                                                          │
    Redact credentials, IPs, tokens before any AI call                   │
         │                                                                │
         ▼                                                                │
    database.py  →  ~/.termstory/termstory.db (SQLite + WAL)            │
    Store sessions, commands, commits, recovery_source, AI cache         │
         │                                                                │
         ▼                                                                │
    tui.py / formatter.py                                                 │
    Render timeline, badges, AI summaries, Chain of Custody tooltips     │

---

5. Shell History Parsing

Zsh — EXTENDED_HISTORY format

When setopt EXTENDED_HISTORY is enabled, Zsh writes:

: 1717500000:3;git commit -m "fix auth"

The parser extracts timestamp, duration, and command text with multiline support (trailing \ continuation).

Zsh — Hybrid / Frankenstein Mode

If you recently enabled EXTENDED_HISTORY, your ~/.zsh_history contains a mix of timestamped and legacy lines. The parser handles this automatically:

• Timestamped lines are extracted normally.
• Legacy lines are collected separately.
• The oldest timestamped command's timestamp minus 60 seconds becomes the anchor_time.
• The Timestamp Detective runs on all legacy lines (see Section 6).
• Timestamp Locking: On subsequent runs, synthetic timestamps are looked up in the database and re-used to prevent legacy commands from shifting dates.

Bash — HISTTIMEFORMAT

If HISTTIMEFORMAT is set, Bash writes #<timestamp> headers before each command. The parser associates each command with its preceding timestamp header. Without headers, timestamps are spaced 10 seconds apart backward from the file's mtime.

Filtering

Commands older than 5 years or with future timestamps are silently dropped to prevent database pollution.

---

6. The Timestamp Detective (v0.2.9)

The most significant feature in TermStory's history. If you've been using your terminal for years without EXTENDED_HISTORY, your shell history has no dates at all — every command appears to have happened "today". The Timestamp Detective reverse-engineers real timestamps by mining your git log, filesystem metadata, and package manager artifacts.

The Problem

Without EXTENDED_HISTORY, a file like Vansh's 847-command history gets anchored to "today" via a 1-second step-back. His March commits, April npm installs, and May docker builds all collapse into one meaningless session labelled today.

The Solution — Three Phases

Phase A — Virtual CWD Tracker

Before running any detector, the engine replays cd, pushd, and popd commands in sequence to compute the virtual working directory at every point in history. This is critical because git commit searches must be scoped to the correct repository — otherwise git commit -m "fix typo" could match a commit from the wrong project.

idx 0:  ls                     → cwd = "~"
idx 1:  cd ~/Projects/myapp    → cwd = "~/Projects/myapp"
idx 2:  npm install            → cwd = "~/Projects/myapp"
idx 3:  git commit -m "init"   → cwd = "~/Projects/myapp"  ← searches THIS repo only
idx 4:  cd ..                  → cwd = "~/Projects"

Supports: cd - (previous dir), relative .. paths, ~ and $HOME expansion, pushd/popd stack.

Phase B — Five Forensic Detectors

Run in priority order. Each returns (unix_timestamp, source_string) or None.

1. Git Commit Matcher ⭐ Highest confidence

Extracts the message from git commit -m "…" and runs git log --all on the CWD-scoped repository. Fuzzy-matches the message using difflib.SequenceMatcher with a threshold of ≥ 0.85 after stripping conventional prefixes (feat:, fix(scope):) and emojis from both sides.

Multi-Repo Collision Trap: The CWD-derived repo is searched first. If Vansh has three repos with git commit -m "fix typo", only the repo he was sitting in when he typed it gets matched. Fallback to other known project paths only if the local repo has no match.

2. File Stat 🟡 Medium confidence

Stats the artifact created by file-creating commands:

Command	Target
touch <file>	<file> — st_birthtime (macOS) or st_mtime
mkdir [-p] <dir>	<dir>
echo/printf/cat > <file>	redirect target
cp <src> <dst>	<dst>
git init [dir]	<dir>/.git
npm init [-y]	cwd/package.json
python -m venv <name>	<name>/bin/activate
cargo init [dir]	<dir>/Cargo.toml
go mod init	cwd/go.mod

touch -t (explicit timestamp set) is excluded. All paths are resolved against the virtual CWD.

3. Package Manager Install Metadata 🟡 Medium confidence

Command	Artifact
brew install <formula>	$(brew --prefix)/Cellar/<formula> mtime
pip install <pkg>	pip dist-info directory mtime
npm install (local)	package-lock.json mtime
npm install -g <pkg>	global node_modules /<pkg> mtime
cargo add/install <crate>	~/.cargo/registry/src/*/*crate* mtime
gem install <gem>	~/.gem/ruby/*/gems/<gem>-* mtime

Both brew --prefix and npm root -g subprocess calls are cached after the first call.

4. Docker Image Inspector 🟡 Medium confidence

For docker build -t <tag> commands, runs:

docker image inspect <tag> --format='{{.Created}}'

Parses the RFC3339 timestamp and returns the exact second the image was built.

5. Venv / Lockfile Sentinels 🟡 Low-medium confidence

Command	Artifact
bundle install	Gemfile.lock mtime
go mod tidy / go get	go.sum mtime
composer install	composer.lock mtime
poetry add/install	poetry.lock mtime
mix deps.get	mix.lock mtime
git clone <url> [dir]	<dir>/.git birthtime
ssh-keygen -f <file>	<file> birthtime

Phase C — Anchor Interpolation Engine ⭐ The Killer Feature

After Phase B, some commands have real timestamps (anchors) and others are still None. For each contiguous gap between two anchors, unresolved commands are distributed linearly:

t_i = t_a + (t_b − t_a) × (i − i_a) / (i_b − i_a)

Example:

[500] npm install express     → 🔍 ANCHOR  Tuesday 14:00  (package.json mtime)
[501] npm start               → 📐 INTERP  Tuesday 14:07  (between anchors)
[502] git commit -m "add express" → 🔍 ANCHOR Tuesday 14:15  (git log)

npm start is mathematically placed at 14:07 — the proportional moment between the two known events. Not a guess, not today — the most likely real time.

• Prefix gap (before first anchor): 1-second step-back from anchor.
• Suffix gap (after last anchor): 1-second step-forward from anchor.
• No anchors at all: original mtime step-back (same as v0.2.8 behaviour).

Chain of Custody — The Trust Badge

Every command that the Detective recovered gets a recovery_source string persisted to the database. In the TUI, it renders as a badge below the command:

💻 Command Timeline:
  • 14:00:03  npm install express
      [🔍 stat: package.json mtime]
  • 14:07:31  npm start
      [🔍 Interpolated (between stat: package.json mtime → git log: myapp@a3f9b2c)]
  • 14:15:44  git commit -m "add express"
      [🔍 git log: myapp@a3f9b2c]

This turns "how the hell did it know that?" into a "holy crap, this app is smart" moment.

TUI Session Labels

Label	Meaning
✨ npm init, express installed (14:00 - 14:45)	Real EXTENDED_HISTORY timestamp
🔍 Recovered Archive (38 cmds • 14:00 - 14:45)	Detective found anchors + interpolated
📦 Legacy Archive (12 recovered cmds)	Zero forensic evidence, fully synthetic

---

7. Project Resolution

project.py maps a working directory to a human-readable project name using a two-pass strategy:

Working Directory: /home/dev/Projects/termstory/tests
         │
         ▼
Walk up directories looking for .git / .hg / .svn
         │
         ▼
Found root: /home/dev/Projects/termstory
         │
         ▼
Read build manifests:
  package.json → "name" field
  Cargo.toml   → [package] name
  setup.py     → name= argument
  pom.xml      → <artifactId>
         │
         ▼
Normalize: strip hyphens, fix casing
Result: "termstory"

Fallback strictness: If a directory is not inside a standard project root (~/Projects/, ~/src/, etc.) and has no VCS markers, it maps to the user's home directory and is grouped under "Other". This prevents ~/.ssh, ~/Downloads, or /tmp from polluting the project list.

---

8. Git Commit Correlation

git_integration.py enriches sessions by fetching matching commits:

git -C <repo_root> log --all \
    --since="<session_start - 5min>" \
    --until="<session_end + 10min>" \
    --format="%H|%at|%s"

Commit cleaning pipeline:

• Strips conventional commit prefixes (feat(scope):, fix:, chore:)
• Removes Unicode emoji and :shorthand: tokens
• Drops merge commit messages and branch pointer refs
• Stores both raw and cleaned message for AI prompts

---

9. Database Schema

~/.termstory/termstory.db — SQLite with WAL mode (PRAGMA journal_mode = WAL).

CREATE TABLE projects (
    id          INTEGER PRIMARY KEY AUTOINCREMENT,
    name        TEXT NOT NULL UNIQUE,
    path        TEXT,
    first_seen  INTEGER,
    last_seen   INTEGER,
    created_at  INTEGER DEFAULT (strftime('%s', 'now'))
);

CREATE TABLE sessions (
    id               INTEGER PRIMARY KEY AUTOINCREMENT,
    start_time       INTEGER NOT NULL,
    end_time         INTEGER NOT NULL,
    duration_seconds INTEGER,
    project_id       INTEGER REFERENCES projects(id),
    ai_summary       TEXT,       -- Cached AI narrative; reused across runs
    created_at       INTEGER DEFAULT (strftime('%s', 'now'))
);

CREATE TABLE commands (
    id              INTEGER PRIMARY KEY AUTOINCREMENT,
    timestamp       INTEGER NOT NULL,
    command         TEXT NOT NULL,
    exit_code       INTEGER,
    session_id      INTEGER REFERENCES sessions(id),
    project_id      INTEGER REFERENCES projects(id),
    recovery_source TEXT,    -- v0.2.9: Chain of Custody attribution string, NULL for real timestamps
    created_at      INTEGER DEFAULT (strftime('%s', 'now'))
);

CREATE TABLE commits (
    hash            TEXT PRIMARY KEY,
    timestamp       INTEGER NOT NULL,
    message         TEXT NOT NULL,
    cleaned_message TEXT NOT NULL,
    project_id      INTEGER REFERENCES projects(id),
    created_at      INTEGER DEFAULT (strftime('%s', 'now'))
);

CREATE TABLE macro_summaries (
    id           INTEGER PRIMARY KEY AUTOINCREMENT,
    timeframe_id TEXT NOT NULL UNIQUE,  -- e.g. "2026-06-03" or "June 2026" or "overall"
    type         TEXT NOT NULL,         -- "date", "month", or "overall"
    summary      TEXT NOT NULL,
    created_at   INTEGER DEFAULT (strftime('%s', 'now'))
);

Indexes: idx_commands_timestamp, idx_commands_session_id, idx_sessions_start_time, idx_sessions_project_id, idx_sessions_date_range, idx_commits_timestamp, idx_commits_project_id, idx_sessions_project_date.

Deduplication: Unique constraints on sessions(start_time) and commands(timestamp, command) prevent re-ingestion duplicates. One-time migration cleans any pre-constraint duplicates on first upgrade.

---

10. Privacy Sanitizer

All data passes through sanitizer.py locally before any AI call. Nothing sensitive ever leaves your machine.

Session Blacklist

If any command in a session matches these patterns, the entire session is short-circuited to return "Security/Authentication Operations" — no commands are sent to the LLM at all:

BLACKLIST_PATTERNS = [
    r'\bvault\b',
    r'\baws\s+configure\b',
    r'\bgh\s+auth\b',
    r'\bkubectl\s+.*?\bcreate\s+secret\b',
]

Redaction Rules

Type	Pattern	Replacement
Private keys	-----BEGIN ... PRIVATE KEY-----	[REDACTED_PRIVATE_KEY]
AWS keys	AKIA[A-Z0-9]{16}	[REDACTED_AWS_KEY]
Bearer tokens	bearer <token>	Bearer [REDACTED_TOKEN]
Flag values	--password, --token, --api-key, -p	--password=[REDACTED]
IPv4/IPv6	standard address patterns	[REDACTED_IP]
Hostnames/FQDNs	host.domain.tld	[REDACTED_HOST]
Exports	export KEY=value	export KEY=[REDACTED]

File extension whitelist: Paths ending in .py, .json, .sh, .yml, .ts, .go, etc. are never redacted even if they look like FQDNs — preserving filenames like config.json and api.ts.

---

11. AI Client

ai.py interfaces with any OpenAI-compatible LLM endpoint using only Python's standard library — no requests, no openai-python.

• Transport: urllib.request.Request with JSON payload
• URL normalization: Strips trailing slashes, auto-appends /chat/completions
• Keyless mode: Skips Authorization: Bearer header if API key is empty (Ollama compatibility)
• Timeout: 15 seconds to prevent blocking the TUI thread
• Background execution: All AI calls run in Textual @work async workers — UI never freezes

Supported Providers

Provider	Default model	Notes
Groq	llama-3.1-8b-instant	Fast, free tier available
OpenAI	gpt-4o-mini	Requires API key
Ollama	llama3	Fully local, no key needed
Custom	any	Any OpenAI-compatible endpoint

---

12. TUI Dashboard

Launch with termstory ui.

┌─────────────────────────────────────────────────────────────────────────────┐
│  🔥 12 day streak  •  43 active days  •  127h 14m total                     │
│  Activity (Last 90 Days):  ░░▒▒░▓▓█▓▒░░▒▒▓▓█▓▒░░▒▓▓█▓▒░░▒▒▓▓█▓▒░░▒░░      │
├───────────────────────────────┬─────────────────────────────────────────────┤
│  June 2026                    │  📂 termstory  •  Tue Jun 03                │
│  ├─ Jun 07 (Sat)              │  ─────────────────────────────────────────  │
│  │  └─ termstory              │  [💻 Dev Log]                               │
│  │     ✨ v0.2.9 Timestamp… │  ├─ 🔨 Built: Timestamp Detective with      │
│  ├─ Jun 06 (Fri)              │         5 forensic detectors & interpolation│
│  │  ├─ termstory              │  ├─ 🔧 Flow: pytest 155 passed, twine      │
│  │  └─ Other                  │         upload, git push origin main        │
│  │     📦 Legacy Archive      │  └─ 🚀 Result: v0.2.9 published to PyPI    │
│  └─ Jun 05 (Thu)              │                                             │
│     └─ 🔍 Recovered Archive   │  💻 Command Timeline:                       │
│                               │  • 00:47:57  git commit -m "feat: v0.2.9…" │
│                               │      [🔍 git log: termstory@bbe9dfc]        │
│                               │  • 00:48:12  python3 -m pytest tests/       │
│                               │  • 00:49:03  twine upload dist/*            │
└───────────────────────────────┴─────────────────────────────────────────────┘
  ?:help  /:search  o:ai-config  c:copy  q:quit

Layout

• StatsHeader (top, full-width): streak count, active days, total duration, and a GitHub-style command-volume heatmap synced to the timeline's --days limit.
• HistoryTree (left 30%): collapsible Year → Month → Day → Project → Session tree. The root node shows an All-Time Wrapped dashboard. Month nodes show Monthly Wrapped.
• DetailsCanvas (right 70%): renders whatever the selected tree node points to — date chronicle, session details, wrapped views, or search results.

Session Tree Labels

✨ npm init, express installed  (14:00 - 14:45)   ← real timestamp
🔍 Recovered Archive (38 cmds • 14:00 - 14:45)    ← Detective recovered
📦 Legacy Archive (12 recovered cmds)              ← fully synthetic

Keyboard Shortcuts

Key	Action
?	Toggle help overlay
/	Open search box (real-time filter of 90-day window)
Enter (in search)	Escape to all-time deep search
Esc	Close modal / clear search / restore timeline
o	Open AI provider configuration screen
c	Copy selected canvas content to OS clipboard
j / k / ↑ / ↓	Navigate tree
Ctrl+↓ / Ctrl+↑	Scroll canvas
q	Quit

Search Engine

Typing in / search filters the pre-loaded 90-day timeline in real-time (no DB hit). Pressing Enter escapes to an all-time database search across commands, commit messages, project names, and AI summaries. The tree root label changes to 🔍 Search Results: "query". Pressing Esc or clearing the box instantly restores the original timeline.

Clipboard

c pipes text directly to pbcopy (macOS), xclip/xsel/wl-copy (Linux), or clip (Windows), bypassing OSC 52 terminal restrictions. ANSI escape codes are stripped before copy.

Onboarding

On first launch with no config, an interactive onboarding screen offers Ctrl+G (Groq), Ctrl+A (OpenAI), Ctrl+L (Ollama), Ctrl+C (Custom), or Ctrl+D (disable AI). If your history lacks timestamps, a consent prompt offers to append setopt EXTENDED_HISTORY to ~/.zshrc.

Wrapped Views

Selecting the root Timeline node or a Month node shows a high-density wrapped dashboard:

• Code churn matrix: git insertions vs. deletions, net LOC growth
• Time distribution: hourly activity punch-card across the period
• Project breakdown: percentage distribution across all projects
• Top commits & tooling: most-used commands, editors, languages
• AI Behavioral Audit: a witty, witty roast of your patterns and productivity archetypes ("The Midnight Alchemist", "The Expansionist Architect") — regeneratable with r

---

13. AI Narrative Design

TermStory's prompts are designed to produce high-density, CLI-styled developer logs — not marketing prose.

Session Summary Format

[💻 Dev Log]
├─ 🔨 Built: <short punchy action — what was built or coded>
├─ 🔧 Flow: <tools used, tests run, configs edited>
└─ 🚀 Result: <milestone shipped, fixed, or pushed>

or:

[🤖 Codebase Pulse]
• Hacked: <what was designed, refactored, or debugged>
• Tooling: <commands run, docker setups, libraries configured>
• Outcome: <what was verified, resolved, or shipped>

Rules enforced in the prompt:

• No paragraphs. No filler. No "ultimately the hard work paid off".
• Every line starts with a past-tense engineering verb: wired up, refactored, debugged, spun up, stabilized, shipped.
• Second-person narrative ("You") for the Daily Chronicle view.

Daily Chronicle (Date node)

When a date is selected in the TUI, a full "Story of You" daily narrative is generated — including an inferred breaks timeline, hourly activity punch-card, and second-person storytelling of the whole day.

---

14. CLI Reference

Daily

termstory                    # Today's timeline
termstory today              # Same as above
termstory today --detailed   # All commands, no noise filtering, with timestamps
termstory today --compare    # Side-by-side with yesterday
termstory today --stats      # Command category frequency table

Search

termstory search <query>
termstory search docker
termstory search docker --project myapp
termstory search docker --since 2026-05-01
termstory search docker --detailed

Historical

termstory week               # Current week
termstory week --last        # Previous week
termstory month              # Current month
termstory month "May 2026"   # Specific month
termstory month --last       # Previous month

Projects

termstory project <name>             # 30-day deep dive
termstory project myapp --files      # Files edited, by frequency
termstory project myapp --stats      # Command category breakdown
termstory projects                   # All tracked projects
termstory projects --sort time       # By total hours (default)
termstory projects --sort recent     # By last active date
termstory projects --sort name       # Alphabetically

Insights

termstory insights           # Focus scores, time-of-day distribution, tool breakdown
termstory insights --days 90 # Over the last 90 days

TUI

termstory ui                 # Full dashboard, last 90 days
termstory ui --days 30       # Limit to last 30 days
termstory ui --all           # All history

Config

termstory config list
termstory config get active_provider
termstory config set active_provider groq
termstory config set providers.groq.api_key gsk_...

Date override

termstory 2026-05-15                 # Specific date
termstory --date 2026-05-15 week     # Week containing that date

---

15. Configuration

Config lives at ~/.termstory/config.json:

{
    "ai_enabled": true,
    "active_provider": "groq",
    "request_timeout_seconds": 30,
    "has_seen_onboarding": true,
    "providers": {
        "groq": {
            "api_key": "gsk_...",
            "api_base_url": "https://api.groq.com/openai/v1",
            "model_name": "llama-3.1-8b-instant"
        },
        "openai": {
            "api_key": "sk-proj-...",
            "api_base_url": "https://api.openai.com/v1",
            "model_name": "gpt-4o-mini"
        },
        "ollama": {
            "api_key": "",
            "api_base_url": "http://localhost:11434/v1",
            "model_name": "llama3"
        },
        "custom": {
            "api_key": "",
            "api_base_url": "http://localhost:8080/v1",
            "model_name": "my-model"
        }
    }
}

All settings are editable via termstory config set <dot.path> <value>.

Key configuration parameters:

• ai_enabled (bool): Toggle AI summaries on/off.
• active_provider (string): Set to "groq", "openai", "ollama", "custom", or "disabled".
• request_timeout_seconds (int): HTTP request timeout (in seconds) for LLM API calls. Defaults to 30.
• providers.<name>.<param>: Provider-specific endpoints, API keys, and model names.

---

16. Testing

python3 -m pytest tests/ -v

v0.2.9 results: 155 passed, 1 skipped, 0 failures

Test File	What it covers
test_parser.py	Zsh format extraction, hybrid mode, bash fill algorithm, multiline commands
test_timestamp_detective.py	Virtual CWD tracker, all 5 detectors, anchor interpolation, full pipeline
test_session.py	30-minute grouping, boundary detection, duration calculation
test_project.py	VCS root climbing, manifest parsing, "Other" fallback, home dir strictness
test_git_integration.py	git log subprocess mocking, commit message cleaning
test_database.py / test_database_queries.py	WAL config, schema init, CRUD, dedup migration, date range queries
test_sanitizer.py	Blacklist short-circuit, credential redaction, FQDN exclusion
test_ai.py	urllib payload construction, keyless mode, timeout, prompt templates
test_tui.py	Textual widget lifecycle, onboarding flow, search, wrapped view, clipboard
test_formatter_rich.py	CLI output layout and Rich markup
test_integration.py	End-to-end ingestion → DB → render

---

17. Troubleshooting

Reset everything

rm -rf ~/.termstory/
termstory ui   # Fresh start, re-runs onboarding

Force a re-ingest

termstory today --detailed

Reads the history file fresh and updates the database.

Enable EXTENDED_HISTORY manually

echo '\nsetopt EXTENDED_HISTORY' >> ~/.zshrc
source ~/.zshrc

Future commands will get real timestamps. The Timestamp Detective handles everything you typed before enabling this.

Local AI (Ollama)

ollama pull llama3
ollama run llama3
termstory ui   # Press 'o' → Ctrl+L to select Ollama

My old history all shows up as "today"

This is expected if you never had EXTENDED_HISTORY. The Timestamp Detective will recover what it can using git commits, file stat, and package manager artifacts. Commands it can't place with evidence are grouped in 📦 Legacy Archive or 🔍 Recovered Archive in the TUI.

The detective didn't recover my history

The Detective needs forensic artifacts — git repos, installed packages, or created files — that still exist on disk. If you've wiped your machine since typing those commands, the artifacts are gone and full recovery isn't possible. Enable EXTENDED_HISTORY now so this doesn't happen going forward.

---

License

MIT © TermStory Contributors

GitHub: https://github.com/bitflicker64/Termstory
PyPI: https://pypi.org/project/termstory/