TwilioWorld Agentic Coding Toolkit

Experimental toolkit for the 2026 Twilio Assemble series. It is actively
developed, may change without notice, and is not a supported production tool.

Give coding agents reliable Twilio context with Twilio Skills and live,
read-only documentation search. The toolkit also offers optional private local
chat, Dev Phone, and an isolated Twilio CLI—all from one terminal dashboard.

Quick start

git clone --recursive https://github.com/andrewdmaclean-work/twilioworld-agentic-coding-toolkit.git
cd twilioworld-agentic-coding-toolkit
./toolkit

First run selects the AI bundle by default:

• Twilio Skills and live Docs search
• Private local chat, opened in the llamafile browser interface or Pi (about 2.5–3.3 GB)

Dev Phone and its account-facing Twilio CLI are additional and remain
unselected. You can deselect the local model during review and configure an
existing coding agent without the large download.

Applying the defaults installs the selected model/runtime and Pi. Opening
Local chat then lets you choose the browser or terminal interface.

If anything looks wrong, run the read-only environment check:

./toolkit doctor

What is included

The toolkit does not give agents authenticated access to your Twilio account.
Live Docs search is remote but read-only and does not require credentials.

Dashboard guide

The main menu is ordered around common tasks:

1. Set up a coding agent — add Twilio Skills and live Docs, then launch it
2. Local chat — choose the simple browser interface or Pi
3. Dev Phone — install or open the browser test phone
4. Twilio CLI — install, sign in, switch account, or open an isolated terminal
5. Resources — open TwilioWorld or Twilio AI documentation
6. Settings — install or repair features, manage local AI, or uninstall/reset

Every action shows a summary before potentially destructive work. Long-running
tasks have progress, detailed output, failure recovery, and retry controls.

Coding-agent setup

Choose Set up a coding agent, review the action, and continue. The toolkit
uses package-manager installers where available, configures Twilio knowledge,
and opens the agent in a new terminal window.

Manual Docs MCP endpoint:

https://mcp.twilio.com/docs

Manual Skills source:

vendor/twilio-ai/skills/

Local AI

Local AI uses llamafile and exposes an
OpenAI-compatible endpoint on 127.0.0.1:8080 by default. Model and runtime
downloads are checksum-verified before use.

Reasoning Off is enforced in both Pi and browser requests, and at the model
server with a zero-token reasoning budget. Low, Medium, and High use increasing
local token budgets; showing Pi reasoning blocks is a separate presentation
setting.

Available models:

Choose Local chat to open the shared model in one of two interfaces:

• Browser chat (llamafile) — a simple built-in chat interface, preconfigured with live Twilio Docs search.
• Pi (terminal) — a coding-focused interface with Twilio Skills and live Docs tools.

Under Settings → Manage local AI you can:

• choose the model shared by the browser interface and Pi;
• disable reasoning or set its effort to Low, Medium, or High independently from whether Pi displays its reasoning blocks (browser chat only receives reasoning when effort is enabled);
• stop the local service; or
• remove downloaded model files.

Set a different port or context size before launching:

MODEL_SERVER_PORT=8082 CTX_SIZE=32768 ./toolkit

Dev Phone safety

Dev Phone can replace a Twilio number's webhook settings. Use a spare test
number, not a production number. The toolkit-local Twilio CLI and its profiles
live inside this repository and do not modify your normal system CLI profile.

Files created locally

Runtime state, credentials, models, logs, and generated configuration are
ignored by Git. Use Settings → Uninstall or reset to review individual
removals, or run the standalone guided uninstaller:

./uninstall.sh

Platform support

Host requirements are bash, git, curl, and standard archive tools. The
toolkit installs project-local Bun and Node runtimes when needed.

Development

cd tui
bun install --frozen-lockfile
bun run typecheck
bun test

cd ..
./test.sh --no-net

The test suite includes OpenTUI headless interaction tests, behavior tests for
configuration and MCP output, integrity checks, TypeScript compilation,
ShellCheck, and a TUI smoke test. See CONTRIBUTING.md for the
development workflow and SECURITY.md for reporting guidance.

Troubleshooting

• The TUI will not start: run ./toolkit doctor.
• Skills are missing: run git submodule update --init --recursive, then
  apply setup again from Settings.
• Local AI does not start: inspect models/pi-server.log, stop local AI in
  Settings, and retry.
• An agent did not configure: review the task details; the toolkit prints
  the agent-specific manual step when its CLI has no stable configuration API.
• A terminal window did not open: launch the agent or Twilio CLI manually
  from this repository so it inherits the toolkit's local paths.

Twilio Skills and Twilio MCP are Public Beta Twilio surfaces and are not covered
by Twilio Support Terms or an SLA.