AgInTiFlow — Local-First Agent Workspace for Real Projects

Web + CLI + API agent workspace

AgInTi Flow local-first agents for real projects

AgInTiFlow is a local-first, project-aware workspace with visible tools, resumable sessions, guarded execution, and explicit hosted-model routes.

$ npm install -g @lazyingart/agintiflow

$ aginti web --port 3210

View real app screenshots Docs GitHub npm

AgInTiFlow terminal launch screen with colorful banner, Docker workspace status, and interactive chat input. — **CLI + Web, one session** Start with `aginti`, then inspect the same project sessions in the browser.

AgInTiFlow web console showing conversation history and structured run output. — **Web console visibility** The browser shows the conversation and run output from the same local workspace.

Routing lane LocalLLM Fast

Local planning and task routing without silently selecting a hosted provider.

Main lane LocalLLM Deep

Substantive local agent work, with hosted routes available only when explicitly chosen.

Safe execution Docker + policy

Read-only by default, explicit approval for setup commands.

Real screenshots from the local app

Daily agent work should feel inspectable, not mysterious.

The carousel uses screenshots captured from the live AgInTiFlow app running at http://127.0.0.1:3210/.

AgInTiFlow web app showing task controls, run output, and conversation history.

Agent cockpit

Route controls, mock run output, wrapper status, sandbox controls, and chat history in one page.

Task and model route controls in the AgInTiFlow web app.

Route and task controls

Pick smart routing, manual provider/model, shell access, command cwd, and package-install policy.

AgInTiFlow run output with structured session logs.

Structured run logs

See every model response, tool call, command result, and final answer as the session runs.

AgInTiFlow conversation history panel after a mock run.

Resume the conversation

Sessions persist chat history so follow-up requests continue from the selected run.

Responsive by default

The same controls remain usable on narrow screens for quick monitoring and setup.

Platform shape

One runner, multiple model and tool routes.

Smart routing

LocalLLM Fast routes work and LocalLLM Deep handles substantive tasks. Hosted providers remain explicit optional routes.

Wrapper tools

Codex, Claude, Gemini, Copilot, and Qwen wrappers are optional advisory tools, not hard dependencies.

Safe file tools

List, read, search, write, and patch workspace files with path guardrails, hashes, and compact diffs.

Mock mode

Local mock routing verifies UI, API, tool execution, and CI smoke paths without DeepSeek or OpenAI credentials.

Runtime documentation

Choose the right execution mode for the job.

AgInTiFlow documents Docker sandboxing, host full-access mode, durable tmux sessions, package installs, and rolling-plan autonomy as first-class product behavior.

Mode guide Docker, host, tmux, and autonomy

When to use read-only, workspace Docker, trusted host access, or future persistent containers.

Self-development Supervised AgInTiFlow-on-AgInTiFlow work

How to let AgInTiFlow edit its own source while Codex supervises scope, checks, git, and release steps.

CLI + web sync Runtime pipe and queues

How CLI and web share sessions, inbox messages, queued prompts, artifacts, and stop/resume state.

Engineering loop Large codebase strategy

Codebase maps, deterministic patching, scout blackboards, checks, and context discipline.

Safety model

Useful automation with explicit boundaries.

AgInTiFlow keeps local commands guarded by an allowlist, stores resumable run history, and exposes sandbox status through APIs that do not leak API keys or npm tokens.

Docker read-only mode for inspection and tests.
Docker workspace-write mode only for approved setup commands.
Workspace edits stay inside the selected command directory and block secret-like paths.
Package installs default to approval-required policy.
NPM publish, token commands, sudo, destructive git, and shell chaining are blocked.

A Web, CLI, and API workspace for agents that work across creative, scientific, software, hardware, and industrial projects with guarded tools, durable sessions, SCS supervision, and AAPS workflows.