Finish Local AI Agent System

Job ID: 40387423

Budget: ₹400 – ₹750 INR

Need a freelancer to finish and harden my AI agent system on macOS

Goal: fully local, no paid API cost for core intelligenceband daily running tasks

Main requirement: 24/7 autonomous local AI assistant for coding, automation, supervision, and future expansion

Current stack already exists:

Mac mini M4, 16 GB RAM

Ollama running locally

local models available:

gemma4:e4b

qwen2.5-coder:14b

deepseek-r1:8b

llava:7b

moondream

nomic-embed-text
Qwen 3.5-9b


Claude Code installed

OpenCode installed

OpenClaw available

n8n available

custom Jarvis v1 controller already created

local dashboard already created

Telegram bot already connected: @bot


What is already partly working:

local Ollama inference works

Telegram /health works

Jarvis controller runs

dashboard runs

launchd services are running

safety blocking exists for dangerous commands

task execution works

registry/workflow creation partially works


What I need finished:

make Claude Code fully usable with local Ollama only

ensure all Claude Code features possible locally are properly configured:

skills

agents

slash commands

hooks

MCP/tool integrations

repo-aware coding workflows

edit/test/fix loops


make sure it never falls back to paid/cloud model usage for core work

make local wrapper/default path foolproof


Core architecture I want:

Jarvis controller = real authority/brain

Claude Code = primary coding/operator interface

Ollama = local inference backend

Telegram bot = remote control interface

dashboard/web UI = monitoring/control

OpenClaw/n8n only if useful, not the main authority


Required capabilities:

natural language prompt → classify → plan → execute → verify → log

create/upgrade:

skills

tools

workflows

agents/teams


local coding and repo automation

shell/terminal execution

file read/write/search/edit

git operations

process/service control

logs + health monitoring

AppleScript/macOS control

screenshot/screen tools

browser automation

Playwright integration

GUI/window/app control if possible locally

future self-upgrade capability through prompts


24/7 stability requirements:

supervisor/watchdog

stuck-task detection

retry logic

duplicate-worker prevention

crash recovery

health endpoints

structured logs

checkpoints / rollback

safe approvals for risky actions


Telegram requirements:

send prompt remotely to bot

controller executes locally

bot replies with result/status

support /health, /status, /tasks, /help

show task ids / failures / summaries

keep Telegram as transport only, not intelligence layer


Main unfinished gaps right now:

Playwright/browser automation not fully installed/wired

Claude local wrapper path needs hardening

Telegram UX needs improvement

resilience and supervision need final polish

OpenClaw sandbox depends on Docker, but this is lower priority


Future expansion I want after

web development skill

Android app builder

tablet control layer

3D modeling/design workflows

business manager automation

customer support automation

repo ingestion / self-upgrade workflows

multi-agent swarm/team workflows


Important constraints:

local-first

no monthly API dependence for core intelligence

prefer free/open-source/local tools

use internet only for:

research

public info

downloading free tools/repos

browser automation tasks


system must be maintainable, not hacked together


What I want delivered:

fully working local stack

clean architecture and file structure

all services running reliably

proper docs: setup, start, stop, recover, use

validated local Claude Code + Ollama workflow

working Playwright/browser automation

working Telegram remote control

safe supervision and recovery

ready-to-use prompt-based upgrade workflow


Freelancer should be strong in:

macOS automation

Python and/or TypeScript

AI agent systems

Ollama / local LLM tooling

terminal automation

Playwright/browser automation

AppleScript/macOS scripting

service supervision / launchd

system architecture / reliability engineering


Ideal outcome:

I can just give prompts in terminal or Telegram and the system handles coding, automation, upgrades, and future capability expansion locally with minimal manual intervention.