Changelog

What's new in Threadrun. The app updates itself; your servers follow it.

0.5.0 · November 9, 2026

Repeat what works.

Turn any thread or factory run into a loop with Repeat…, or pick one on the Loops page. The loop gets exactly what it had: every message, attached files, agent, model and settings, skills & MCP, stages and project.
Updates never interrupt a working agent: they wait until it's done, and tell you so.
Write longer tasks: the message, factory goal and loop task boxes take several lines (Shift+Enter for a new line).
A calmer thread header: the title, status and context are always readable; less-used actions are under More actions.
The context meter keeps its last real value instead of dropping to 0%, and fills in when a provider doesn't report it.

Filter models in the picker: free, smartest, cheap, long context, reasoning, tools, vision and open weights. Combine them to find, say, strong models that cost little.
Intelligence ratings no longer go missing when the leaderboard is briefly unreachable, and filters that would show nothing are hidden.

The model picker's Cancel button keeps its size; where the model data comes from is now at the bottom.

Choose models with confidence.

The model picker shows each model's intelligence (one to five dots), cost, context size, and what it can do: reasoning, tools, vision.
Free models show their limits, so you know how far they go before you start.
Type free in the picker to see only the free models.

A built-in browser for HTML.

HTML files the agent makes open as real pages inside Threadrun, with their styles, scripts and images. Switch to Code to read the source.
Links to other websites open in your own browser.

Find anything.

Phone and website fixes.

On your phone, permission requests show what the agent wants to run, and update when they are answered on your computer or by your policy. They no longer stay open after the work is done.
The website fits phone screens: the header keeps Docs and Download in view.

Previews everywhere.

Open in browser and report previews work when your threads run on your own server.
The phone has a Files tab: open the pages, PDFs and images the agent made, right on your phone.

Paste long text as an attachment.

Pasting three or more lines, or a long passage, into a message attaches it as Pasted text instead of filling the box. You see a card with its first lines and its length, and can remove it before sending. The agent gets the full text with your message.
Works the same on your phone.
Files you attach to the first message of a new thread now reach the agent.

Threadrun keeps itself up to date.

The app checks for new versions, downloads them in the background and verifies them. Click Restart to update when it suits you. See Install.
Your servers follow the app: when it updates, each server's runner updates too. You can turn this off per server. See Your own server.
Messages read the way agents write them: lists, tables, code blocks and links on your computer and your phone. JSON is pretty-printed, diffs are colored, and file paths open in Files.
Clean up on a server card frees disk space and memory, and Stop now stops every agent on that server. Each button explains what it does.
Remove MCP servers and skills you added, including ones from Discover.
Updating a server needs under 1 GB of free disk.
The docs are new: everything from installing to factories, loops and governance.

Run agents on your own server. Free.

Connect any Ubuntu server you can reach over SSH. Threadrun checks it, shows exactly what it will install, and sets it up after you approve. Agents keep working while your laptop sleeps.
Follow and answer your agents from your phone. Scan a code and enter the PIN only your app shows.
Switch runners from the top bar, and choose where each factory and loop runs.
Agents can search the web, and MCP servers that need a key ask for one.
Small servers work too, one agent at a time.
See exactly what a thread changed, file by file, with diffs.
Keep long conversations healthy: a context meter, Compact and Clear.
Threadrun is free: no account, no subscription.

Loops, and runs that get cheaper.

Loops run agents on a schedule, on a webhook or continuously, and stay quiet until something needs you.
Every run ends with a review: what it cost, where the tokens went, and how to make the next run cheaper, with proof on the next run.
Browse each thread's files, preview them, and copy the work to your computer.
Attach files and images to messages and factory goals.
Discover MCP servers and skills from online directories, and choose them per agent.
Find any model quickly with a searchable picker.
Idle threads are archived to save space and wake up on your next message.

Factories: a team of agents on one goal.

Chain agents into a pipeline, or let a lead agent build its own team.
Pick the agent, model and effort for each stage, with approval gates and budgets.
Watch every message between agents as it happens.
Governance: tool rules, network allowlists, budgets, roles, approvals and a tamper-evident audit log.

The first Threadrun.

Run Claude Code, Codex and OpenCode in isolated workspaces, each with its own copy of your repository.
Follow every step, answer permission requests, and see tokens and cost as you go.