Runs automatically
A goblin doesn't wait to be asked. It wakes on a schedule or on events from GitHub, GitLab, Slack, and Linear — the work is underway before anyone typed a prompt.
One loop
Six moving parts
Every AI goblin runs the same loop — discover, build, verify, remember. Here's the machinery underneath.
Pulls in your context
Before touching code, a goblin digs through your issues, threads, docs, and history — so it acts on the whole picture, not just the diff in front of it.
Works in isolation
Every goblin works in its own isolated cloud sandbox — in parallel, around the clock. Your credentials are never exposed to the LLM.
Checks its own work
The goblin that wrote the code never grades its own work. A second one runs an adversarial review using a different LLM provider, reconciling findings in a loop until nothing's left dangling.
Follows your conventions
Your repos, your test commands, your code conventions — kept in a shared playbook every goblin reads. Correct one once, and every goblin after it gets it right.
Learns from past runs
Every shift ends with a handoff note — what worked, what broke, what's next. The next goblin starts where the last one stopped, never from zero.
Demo video — coming soon
Demo video — coming soon
Every pull request
reviewed Automatically
The moment a pull request is opened, a goblin clones the repository, gathers the context, analyses the changes using frontier models, and publishes a detailed review — a summary, security findings, and concrete improvement suggestions as inline comments.
Mention @taskgoblin to ask follow-ups or have it implement the fixes for you.
Hand off an issue, get a pull request back
Assign an issue to @taskgoblin — just like you would to a teammate — and get a pull request ready for review in minutes, so you can focus on the decisions that matter.
Ask @taskgoblin to plan first and it replies with a “Coding Plan” that breaks the work into manageable steps. And there's no limit on how many you can hand off at once — goblins work in parallel, each in its own isolated cloud sandbox, clearing your entire backlog in minutes.
Demo video — coming soon
Demo video — coming soon
Demo video — coming soon
Demo video — coming soon
Demo video — coming soon
Demo video — coming soon
Context is what it knows.
The harness is how it works.
The loop is what keeps it going.
A prompt gets you one answer. A loop gets you a teammate: goblins wake on schedules and events, do the work, check it, write down what happened, and pick it up again on the next turn — no one standing by to press enter.
You don't have to start from a blank page. Pick one of our many templates — overnight triage, review every pull request, patch what monitoring flags — and your first loop is running today.
FAQ
What is TaskGoblin?
TaskGoblin is built for agentic SDLC workflows. It pulls your org's operating context together from across every tool and team conversation — code, tickets, docs, monitoring, cloud — and reasons over it to move work forward in the channel or thread where it's already happening. It investigates, plans, opens PRs, captures team knowledge, and keeps everyone aligned as work evolves.
What is Agentic SDLC?
Agentic SDLC is a software delivery practice where AI agents participate meaningfully across the complete software development lifecycle — planning, requirement analysis, design, coding, testing, deployment, and maintenance. Instead of your team bouncing between point tools, our goblins work alongside you, carrying context across phases, taking action on your behalf, capturing what the team learns, and staying accountable for its actions.
Can goblins work on several tasks at once?
Yes. Every task runs in its own branch and sandbox, so they never step on each other. Hand over five issues at once and the goblins work all five in parallel, opening a separate merge request for each — no queue to babysit.
Which tools does TaskGoblin connect to?
Slack, GitHub, GitLab, and Linear today — each one is a 2-click connection. More connections are on the way.
How do I get started?
Sign up for free, in 2 only clicks, and hand over your first task — assign an issue to the goblin or mention @taskgoblin in a thread. No credit card required.
How does pricing work?
You pay for time, not tokens. There are no plans and no subscription — you buy prepaid credits, and each run is metered per second of the agent's actual working time, billed at $0.40 a minute. Bring your own LLM account and it drops to $0.20; bring your own LLM account and your own sandbox server and it drops to $0.10. Every new organisation starts with $100 in free credits, no card required.
Is my code safe?
Yes. Each goblin works in an isolated sandbox scoped to a single organisation, so one team's work can never touch another's. Goblins only see the workspaces, repositories, and projects you explicitly connect — nothing else — and your credentials are injected around the agent, never handed to it to read. Disconnect a connection at any time and the goblin loses access on the spot.
Do goblins replace my engineers?
No — goblins amplify your team, they don't replace it. They take the repetitive work (reviews, triage, routine fixes) off your plate so your engineers can spend their time on the decisions that actually need a human.
Create your first loop today
Start for free2-click setup · $100 in free credits · No credit card required
Recent highlights
View all posts
Security
How TaskGoblin keeps your secrets out of the agent
Our agent reads untrusted code and clones whole repositories. Here is the proxy architecture that injects secrets into requests without the agent ever holding them.
Engineering
An agent your whole team can @-mention
Personal agent tools made one developer faster. The next interface is not a better window on your laptop — it is an agent that lives in your issue tracker, your repos, and your Slack, where the whole team already works.
Engineering
The work that starts after the merge
Coding agents made shipping cheap, but the systems around your code — issues, docs, CI, release notes — still get maintained by hand. Here is how we close that gap with recurring, autonomous agent runs.