intake

README

         ╔═══╗           █████ █   █ █████  ███  █   █ █████
    ╔════╩═══╩════╗        █   ██  █   █   █   █ █  █  █    
    ║  ─────────  ║        █   █ █ █   █   █████ ███   ████ 
    ╠═════════════╣        █   █  ██   █   █   █ █  █  █    
    ║ q w e r t y ║      █████ █   █   █   █   █ █   █ █████
    ║   a s d f   ║
    ╠═════════════╣      ━━━━━━━━━━━━━ ISSUE FORGE ━━━━━━━━━━━━━
    ║ [  space  ] ║      Ollama · GitHub CLI · Bubble Tea
    ╚═════════════╝      paste anything → structured issue,
    ═════════════════    one command. MIT · local AI · no cloud.

Intake is a terminal UI for GitHub issues powered by a local AI agent. Paste anything — a Slack message, an error log, meeting notes, a stack trace — and intake drafts a structured GitHub issue using a local Ollama model. No cloud. No account. No telemetry.

Status: Early release (v0.1.0). MIT-licensed. macOS and Linux. Report vulnerabilities via SECURITY.md.

---

░▒▓█ TL;DR

intake

Two-pane TUI: browse your filesystem left, manage GitHub issues right. Press c to create an issue — paste anything, the AI agent handles the rest.

---

░▒▓█ Installation script

One command. Installs Ollama, GitHub CLI, the intake binary, pulls the AI model, signs you in, and asks which repo to use.

curl -fsSL https://raw.githubusercontent.com/tinydarkforge/Intake/main/scripts/install.sh | bash

What it does, in plain English:

1. Checks if you have Ollama (the local AI). Installs it if not.
2. Checks if you have GitHub CLI (gh). Installs it if not, then signs you in.
3. Downloads the intake app for your computer (macOS or Linux).
4. Pulls a default AI model (llama3, ~4.7 GB — first time only).
5. Asks which GitHub repo to point intake at, and saves it.

When it finishes, type intake and press Enter.

Re-running the script is safe — every step skips itself if already done.

---

░▒▓█ What it does today

intake wraps two things: a Norton Commander-style filesystem + issues browser and an agentic issue drafter powered by Ollama.

• Two-pane layout — local filesystem left, GitHub issues right
• Agentic issue creation — paste raw context; agent asks 1–3 questions if needed, then drafts from your template
• Template-aware — reads .github/ISSUE_TEMPLATES/*.md frontmatter; auto-applies labels on create
• Full issue CRUD — create, edit title+body, comment, assign yourself, close, reopen
• Streaming drafts — token-by-token output while the model thinks
• No cloud dependency — Ollama runs locally; gh handles GitHub auth

intake does not ship its own AI model. Every draft originates from whatever Ollama model you choose.

---

░▒▓█ Positioning

intake is not a project management tool, a GitHub web app replacement, or a SaaS product. It is a local TUI gate for the create-and-triage loop: go from messy context to a structured GitHub issue without leaving the terminal.

Alternative	When to pick it instead of intake
GitHub web UI	You prefer a browser, or need labels/milestones/projects UI
gh issue create	You already have a clean title and body
Linear / Jira	You need a managed tracker with workflows, not raw GitHub issues

intake's niche: local AI, no account, no telemetry, instant context-to-issue from the terminal.

---

░▒▓█ Requirements

Tool	Purpose	Min version
Ollama	Runs the AI model locally	0.1.x
GitHub CLI (gh)	Creates / lists issues	2.x
Go	Build from source only	1.26+

Platforms: macOS (Intel + Apple Silicon), Linux (x86_64 + arm64), Windows (x86_64 + arm64, untested). Pre-built binaries on the releases page. The guided installer (install.sh) covers macOS and Linux only; Windows users should download the zip from releases.

---

░▒▓█ Manual install (advanced)

Most users should use the Installation script above. The steps below are for anyone who wants control over each piece.

Pre-built binary

Grab the tarball for your platform from the latest release, extract, and move the binary onto your $PATH:

# example — macOS Apple Silicon
tar -xzf intake_*_macos_arm64.tar.gz
sudo mv intake /usr/local/bin/

Builds are provided for macos_arm64, macos_x86_64, linux_arm64, and linux_x86_64.

go install

go install github.com/tinydarkforge/intake@latest

Build from source

git clone https://github.com/tinydarkforge/intake.git
cd intake
go build -o intake .
sudo mv intake /usr/local/bin/

Confirm /usr/local/bin is on your $PATH

echo "$PATH" | tr ':' '\n' | grep -x /usr/local/bin || \
  echo 'export PATH="/usr/local/bin:$PATH"' >> ~/.zshrc

---

░▒▓█ Dependencies

Ollama

macOS / Linux:

curl -fsSL https://ollama.ai/install.sh | sh
# or: brew install ollama

Start the daemon:

ollama serve

Pull the default model:

ollama pull llama3

Other models work — mistral, llama3.1, gemma2. Change the model in the Settings screen.

Verify:

curl http://localhost:11434/api/tags

GitHub CLI

macOS: brew install gh

Linux (Debian/Ubuntu):

(type -p wget >/dev/null || (sudo apt update && sudo apt-get install wget -y)) \
  && sudo mkdir -p -m 755 /etc/apt/keyrings \
  && wget -qO- https://cli.github.com/packages/githubcli-archive-keyring.gpg \
     | sudo tee /etc/apt/keyrings/githubcli-archive-keyring.gpg > /dev/null \
  && echo "deb [arch=$(dpkg --print-architecture) signed-by=/etc/apt/keyrings/githubcli-archive-keyring.gpg] https://cli.github.com/packages stable main" \
     | sudo tee /etc/apt/sources.list.d/github-cli.list > /dev/null \
  && sudo apt update && sudo apt install gh -y

Authenticate:

gh auth login

---

░▒▓█ Configure

Option A — Settings screen (recommended)

Run intake, press s, fill in repo and model, press ctrl+s to save.

Config is stored at ~/.config/intake/config.json (or $INTAKE_CONFIG).

Option B — Environment variables

export INTAKE_REPO="owner/repo"
export INTAKE_MODEL="llama3"
export OLLAMA_HOST="http://localhost:11434"
export INTAKE_TIMEOUT=120
export INTAKE_MAX_TURNS=6
export INTAKE_TEMPLATE_DIR=".github/ISSUE_TEMPLATES"
export INTAKE_DEBUG=1

Option C — CLI flags

intake -repo owner/repo -model mistral -ollama-host http://192.168.1.10:11434
intake -no-sound -debug

Flags override env vars, which override the config file.

---

░▒▓█ Run

# make sure Ollama is running first
ollama serve &

intake
# or with explicit repo
intake -repo owner/repo

---

░▒▓█ Usage

Main view — Norton Commander layout

intake opens in a two-pane layout:

• Left pane — local filesystem browser, rooted at your working directory
• Right pane — GitHub issues list for your configured repo

Press Tab to switch panes. The active pane has a brighter border and title.

Global keys

Key	Action
Tab	switch active pane
c	open Create issue screen
s	open Settings screen
?	show help
q	quit
esc	go back / cancel

Left pane — filesystem

Key	Action
↑ / k	move up
↓ / j	move down
enter	enter directory

Right pane — issues list

Key	Action
↑ / k	move up
↓ / j	move down
enter	open issue detail
o	toggle open / closed
r	refresh list

---

Create issue (c)

Step 1 — Choose template

Pick from .github/ISSUE_TEMPLATES with j/k, confirm with enter.

Step 2 — Title (optional)

Type a short title, or leave blank — the agent infers one from your context.

Step 3 — Context (paste anything)

This is the power step. Paste whatever you have:

• A Slack message: "login redirect broken on staging, users get 403 after SSO"
• An error log or stack trace
• A PR description you want converted to a task
• Meeting notes: "discussed: dark mode, auth timeout issue, mobile perf"
• A one-liner: "fix the thing that breaks when you click save twice"

Press ctrl+s when ready.

Tip: pastes longer than ~150 characters skip the Q&A round entirely — the agent drafts immediately from your context.

Step 4 — Agent Q&A (short pastes only)

For brief inputs the agent may ask 1–3 clarifying questions. Answer and press enter. Type skip to force an immediate draft.

Step 5 — Preview

Key	Action
y	create the GitHub issue
r	regenerate
esc	cancel

---

Issue detail (enter)

Key	Action
↑ / k	scroll up
↓ / j	scroll down
e	edit title and body
c	add a comment
a	assign yourself
x	close issue
o	reopen issue
esc	back to list

Editing: press e, use tab to switch between title and body, ctrl+s to save. Changes are applied with gh issue edit and the issue reloads automatically.

---

Settings (s)

Field	Description
GitHub Repo	owner/repo — where issues are created
Ollama Model	any model you have pulled (llama3, mistral, gemma2…)
Ollama Host	default http://localhost:11434 — change for remote Ollama
Timeout (sec)	how long to wait for the model before giving up
Max Turns	max Q&A rounds (default 6)
Sound	audio feedback on macOS/Linux
Debug	dumps raw model output to stderr

Navigate with tab / shift+tab, press ctrl+s to save.

---

░▒▓█ Issue templates

intake reads templates from .github/ISSUE_TEMPLATES/*.md. Each file needs YAML frontmatter:

---
name: "🐞 Bug Report"
about: "Report an issue"
title: "🐞 [Brief title]"
labels: ["bug"]
assignees: []
---

## Summary
...

Labels are auto-applied on create. Any repo already using GitHub issue templates works out of the box.

---

░▒▓█ Troubleshooting

ollama is not running — start it with: ollama serve

ollama serve

gh: command not found Install the GitHub CLI (above) and run gh auth login.

Model produces garbled JSON Try a larger model: ollama pull llama3.1, then set it in Settings. Small models (< 7B) sometimes struggle with strict JSON output.

Templates not showing Ensure INTAKE_TEMPLATE_DIR points to a directory with *.md files that have YAML frontmatter. Default is .github/ISSUE_TEMPLATES relative to where you run intake.

Sound not working on Linux Install pulseaudio-utils (paplay) or alsa-utils (aplay). Or toggle Sound off in Settings.

---

░▒▓█ Env reference

Variable	Default	Description
INTAKE_REPO	(unset)	Target GitHub repo, e.g. owner/repo
INTAKE_MODEL	llama3	Ollama model
OLLAMA_HOST	http://localhost:11434	Ollama daemon URL
INTAKE_TIMEOUT	120	Request timeout in seconds
INTAKE_MAX_TURNS	6	Max agent Q&A rounds
INTAKE_TEMPLATE_DIR	.github/ISSUE_TEMPLATES	Template directory
INTAKE_DEBUG	(unset)	Set to 1 to log raw model output
NO_COLOR	(unset)	Disable ANSI colors
INTAKE_CONFIG	~/.config/intake/config.json	Override config file path