Agents
No lock-in. Use the agent you already pay for.
Sign in with Claude or ChatGPT and every task runs on that subscription — no API key, no metered credits, no second bill for tokens. Each task gets a fresh microVM, and the credentials are attached from outside it.
- Runs on your Claude Pro or Max plan, or your ChatGPT Plus or Pro plan
- A fresh Firecracker microVM per task, destroyed when the task ends
- Credentials attached by a proxy outside the machine — no token inside the box running your code
- Switch agent per task; PostHog Code is supported too
Free plan: connect as many providers as you like — the cap is on tasks running at once, never on agents connected.
step 1 / 3
Connect your repos
Sign in with GitHub and pick the repos you live in.
You are already paying for a coding agent
Most tools in this space resell you inference. You connect an API key, or you buy their credits, and every run is a second bill on top of the Claude or ChatGPT subscription you already have sitting idle most of the day.
Talyn conducts agents rather than replacing them. You sign in with Claude or with ChatGPT, and the work runs on that plan. There is one flat price for Talyn and nothing metered on top of it.
step 1 / 3
Connect your repos
Sign in with GitHub and pick the repos you live in.
Talyn Fleet: where the work happens
Claude and Codex tasks run on Talyn Fleet — Firecracker microVMs on our own hardware. Every task gets a fresh machine, and the machine is destroyed when the task ends.
The credential handling is the part worth understanding, because it is the bit that is genuinely different. Your subscription token is never placed inside the sandbox. The machine talks to a proxy, and the proxy attaches the credential on the way out. So an agent that reads a hostile repository and decides to exfiltrate your token has nothing in its environment to find.
The egress route table is built per task from the model you picked, which means a Codex run has no network route to Anthropic's API at all, and vice versa. Not a policy — no route.
Picking an agent is picking a model
There is no separate 'which vendor' field, deliberately. You choose a model, and the model determines the vendor, the credential and the routing. A second field would be a second source of truth that could disagree with the first.
You can set a workspace default, or choose 'ask every time' and pick per task from the row itself.
- Claude — on your Claude Pro or Max plan
- Codex — on your ChatGPT Plus or Pro plan
- PostHog Code — in your own PostHog project
What happens when a plan runs out
A monthly quota that is spent is a different problem from a credential that is broken, and Talyn treats them differently. A revoked or expired grant fails visibly and asks you to reconnect, because somebody who chose to run on their own subscription should not be quietly moved onto metered credits to fix it.
An exhausted quota is just the month running out, and the work is still wanted — so that run moves to your other subscription if you have one, and then to PostHog Code. Each hop is recorded on the task. The exhaustion is remembered so the next task does not boot a machine to discover the same thing, and the hold clears on the vendor's own stated reset time, a successful run, or a fresh sign-in.
What we do not offer
Claude Code is not a provider. It was one, and it was removed: it billed metered API credits, which is the opposite of what this page is about. Running Claude on the fleet, on your own subscription, is what replaced it.
Codex Cloud is deferred — OpenAI ships no server-to-server API for it, only a CLI and a GitHub mention flow. Running Codex on the fleet works and is a different thing.
And the fleet is finite hardware. It paces itself, and it spills to the fall-back provider when it is full, but nobody should read this page as a promise of unlimited parallelism.
Questions, answered.
You bring your own. Sign in with Claude or ChatGPT and Talyn Fleet runs your tasks on that subscription — no separate API bill. PostHog Code is supported too. Talyn conducts whichever one you connect, and you can switch per task.
On Talyn Fleet, each task runs in its own Firecracker microVM on our hardware, using your agent subscription. The VM reaches your repository and your agent through a proxy that attaches the credentials from outside, so no token is ever inside the machine running the code, and the VM is destroyed when the task ends. With PostHog Code the run happens in their cloud under your account instead.
No. You sign in with Claude or with ChatGPT and Talyn uses that subscription. An API key would mean metered credits, which is the bill this is designed to avoid.
Talyn talks to GitHub and your chosen provider over their official APIs using credentials you supply. The run happens in a sandbox that is destroyed afterwards, your subscription token is never inside it, and Talyn never stores your source.
Talyn does this for you.
Mission control for your GitHub pull requests, running on the Claude or ChatGPT subscription you already pay for. Free for three tasks at a time.
Keep reading
Feature
Fix a PR with an agent
Failing tests, conflicts, review comments
Read moreFeature
MCP servers
Linear, Sentry, Supabase, your own
Read moreFeature
Loops
Prompts on a schedule
Read moreCompare
Talyn vs Codex cloud
Codex cloud runs tasks in an isolated environment on your ChatGPT plan and opens a pull request. Talyn runs Codex on that same plan, but its unit is the pull requests you already have rather than the task you dispatched.
Read the comparisonCompare
Talyn vs Cursor cloud agents
Cursor's cloud agents run tasks in a VM and hand you a merge-ready pull request, billed at API pricing. Talyn manages the pull requests you already have and runs its agents on the subscription you already pay for.
Read the comparisonCompare
Talyn vs Devin
Devin is an autonomous engineer you assign work to. Talyn is mission control for the pull requests you already have. One writes the change; the other gets it merged. Where each fits.
Read the comparison