Local-first agent runtime

AgenticLocal

A small, inspectable agent loop with tools, workspace policy, durable SQLite state, learning drafts, streaming events, voice, and runtime-selectable model adapters.

Try it locally

Install and start exploring

curl -fsSL https://raw.githubusercontent.com/Bowen-AI/AgenticLocal/main/scripts/install.sh \
  | bash -s -- --with-ollama
agentic-loop chat --enable-network-tools

What it looks like

CLI, voice, and event streams on the same runtime

Terminal chat using network tools

Terminal chat

Continuous local chat with file, memory, date/time, web, news, and fetch tools.

Event stream and registry views

Streaming events

Run, model, tool, approval, memory, and final-answer events persisted to SQLite.

Architecture

Built around the real agent loop

Context Builder
Brain Adapter
Policy Engine
Tool Registry
Event Logger
SQLite Storage

The model chooses an action. The runtime checks policy, executes approved tools, records events, persists state, and feeds the observation back into the next step.

  • Named read and write roots with path escape protection.
  • Durable sessions, memory, traces, rules, workflows, and registry metadata.
  • Tool-call loop guard for repeated identical successful calls.
  • Rule, Ollama, OpenAI-compatible, Gemini-compatible, and LocalAI adapter paths.
  • Provider/model selection per run, chat request, or session.

Use cases

Practical local workflows

Workspace assistant

Inspect files, read notes, and write safe outputs under configured roots.

Dataset triage

Preview CSV columns, row counts, missing values, and numeric ranges.

Research and news

Opt into web search, Google News RSS search, and URL fetching.

Voice companion

Use the browser voice page now; plug in realtime voice adapters later.

About us

Built by Bowen AI

AgenticLocal is created and maintained by Bowen. Learn more about the builder and background on the personal page.

Docs

Everything needed to show the project