In partnership with
I keep coming back to the same math this week: every dollar I send to a hosted API is a dollar I'm choosing not to save. Ollama crossed another star milestone and it's still the fastest path from a bare machine to a running local model, no account, no rate limit, no invoice. Pair it with a real multi-agent harness and you've got a stack that doesn't care what OpenAI charges this month. That's the frame for today: what can you run yourself, right now, for free.
The AIgent stays free for everyone, and reader support is what keeps it that way. Show your support for The AIgent so we can keep providing for and building the community together.
|
Sponsored
|
Your customer messaged on Instagram. You never saw it.
Your customers reach out on the channels they already use — Instagram, Facebook Messenger, WhatsApp, SMS — and when no one's there to answer, they move on to a business that was.
Wati connects those channels into one inbox with AI-powered automations that help you reply the moment a message lands. So you show up where your customers already are — and never leave them waiting.
The Drops
Repoollama
178,896 stars · ollama/ollama
Run any open-source LLM locally, no cloud account required.
Here is the real problem it solves. Every hosted API call is a recurring cost that scales with usage, and most agent workflows don't need frontier intelligence for every step. Ollama gives you a local inference server you control completely.
Quick start: ollama run llama3 and you have a model answering in your terminal in under two minutes.
Catch: local models still lag frontier models on complex reasoning, so keep the hard steps routed to a hosted model and offload the rest.
AffiliateMurf AI
The one piece of this week's run-it-yourself stack you'll still want hosted: the voice.
Murf AI, a text-to-speech studio with realistic voices in many languages and control over tone, pacing, and emphasis. It also ships a low-latency voice API and SDKs, so you can wire speech into an agent or app, not just export a voiceover.
Use it for: giving an agent a low-latency voice with tone and pacing you control.
We may earn a commission.
Repomunder-difflin
1,935 stars · chaitanyagiri/munder-difflin
A local multi-agent harness built to run entirely on your own machine.
Use it for: pairing with Ollama to run a full agent crew without a single API key in your env file.
Catch: young repo, expect rough edges in the orchestration layer.
Repocline
66,420 stars · cline/cline
An autonomous coding agent that runs as an SDK, IDE extension, or CLI.
Use it for: dropping into VS Code today if you want an agent that edits files and runs commands without leaving your editor.
Catch: give it guardrails on file scope before you let it loose on a real repo.
Repoai-memory
2,637 stars · akitaonrails/ai-memory
A long-term memory solution built for agent coding CLIs.
Use it for: handing context between different agent vendors without re-explaining your whole project every session.
Catch: you still have to decide what's worth remembering, the tool won't curate for you.
|
From Our Partners
|
Find the perfect marketing agency, for free.
Stop writing RFPs. Tell us your budget and needs, and our experts send back 3 or 4 agencies worth pitching. The no cost way to source the best agency, without lifting a fingers
Start Here
Give it a role and one example, not a question.
Most people type a question into ChatGPT the way they'd ask a coworker, and get back something generic because the model has no idea who it's supposed to be or what "good" looks like to you. Fix both in one prompt.
1. Open any chat window you already have.
2. Tell it who it is: "You are a blunt copy editor who hates filler words."
3. Show it one example of the output you want, even a single sentence you like.
4. Then ask your real question.
5. Compare that answer to what you'd have gotten from just asking the question cold, you'll notice the difference immediately.
TRY THIS: Take the last question you asked an AI chatbot, add "You are a [specific role]" and one example sentence of the tone you want, and ask it again.
|
Recommended
Once you've got the role-and-example habit down, the gap between you and the people getting scary-good results is just reps: a little structured practice every day beats another folder of saved prompts. Use AI like the top 1% in 10 min/day is the whole pitch: 25,000+ pros already use it, there are 350+ tutorials with new ones landing every week, and you can cancel anytime. We may earn a commission. |
Frontier Signals
Microsoft Copilot leaked passwords through a hidden input. A secret parameter let attackers steal credentials the moment a target clicked a link, no malware needed. If you're routing any customer-facing workflow through Copilot, audit what inputs it actually exposes before you trust it with anything sensitive. (Ars Technica AI)
OpenAI paused a chunk of its training runs after Astra showed signs of "critical" cyber capability. That's not a marketing pause, that's a lab admitting its own model got dangerous enough to stop and rebuild the safety net first. Worth watching before you build anything on Astra's back. (Wired AI)
OpenAI tightened its own monitoring after a Hugging Face breach touched its development pipeline. More detailed model monitoring during training and post-training now ships as policy, not just PR. If you fine-tune on shared infrastructure, assume the bar just moved. (TechCrunch AI)
Firefox's Smart Window can now pull live web results into AI chats and show its sources inline. A partnership with Exa means the browser itself is becoming an agent with citations, not just a portal to one. (The Verge AI)
Qwen 3.8 27B scored 52 on the Artificial Analysis Intelligence Index, tying GPT-5.6 Luna and landing one point behind GLM-5.2. A 27B open model matching a flagship-class score is the exact kind of cost-floor collapse worth checking against whatever you're currently paying per token. (Simon Willison)
|
Recommended reading
If you like The AIgent, a small group of operator-tier publications worth your inbox: see the shortlist. |
Which best describes you right now?
Before You Go
What are you running locally that used to cost you an API bill? Reply and tell me, I read every one and I'm always looking for the next thing to try before I write about it.
See you Thursday.
|
Want to reach builders shipping with AI every weekday? Advertise in The AIgent. |
