Run a coding agent in the cloud.
A coding agent is only as useful as the computer under it. plori gives you a coding agent with a real one: a shell with git and dev tools, a persistent disk for working copies, and enough runway to finish the build after you close the tab.
Opens your agent straight away, no signup needed.
A real dev loop, not a snippet runner.
§01Clones and keeps repos
Working copies live on the agent's persistent disk, so follow-up asks skip the setup and start from where it left off.
Installs its own toolchain
Node, Go, Python, whatever the repo needs. Installs persist with the agent instead of evaporating per session.
Runs builds and tests
Long compiles and suites run as background tasks and report back in-thread, instead of timing out a chat turn.
Asks before it leaps
Credentials and risky commands go through built-in human-in-the-loop: the run parks, you approve or supply the secret (masked), it resumes.
Remembers your conventions
Memory notes carry your stack, style, and standing constraints into every turn.
Delegable over MCP
Hand work to it from Claude, your IDE, or CI via the MCP server, then poll the run or get the reply in-thread.
Where it fits next to your IDE.
§02An IDE assistant is at its best in the tight loop: you watching, it editing. A hosted coding agent covers the other half of the job, the part that does not need you present: the long test run, the overnight refactor, the "check CI in the morning and fix what broke" standing order via scheduled runs.
Because plori speaks MCP in both directions, the two compose: your IDE agent can delegate to your plori agent, and your plori agent can be steered from anywhere. Start one at plori.ai/agent with no signup and hand it a repo.
Questions, answered.
§03What can a cloud coding agent actually do?
Everything that fits in a shell: clone repositories, read and edit code, install toolchains, run builds and test suites, use git, and write up what it did. On plori it does this on its own computer with a persistent disk, so the repo it cloned Monday is still there Friday.
How does it authenticate to GitHub or other services?
It asks you. When a command needs a credential, plori's human-in-the-loop flow parks the run and requests the value from you with a masked input, so secrets are redacted from the visible history and never guessed by the model.
What happens on long builds or test runs?
They move to the background instead of hanging the conversation. The agent answers you, the job keeps running on its machine for up to about an hour, and the result posts back into the same thread when it finishes.
Which models does it code with?
You do not choose one. The Plori Router picks a model per task, the cheapest that can do the job well. Free routes among fast cost-efficient coders (DeepSeek V4 Flash, GLM 5.2, GPT-5.6 Luna); paid plans add the frontier models (Claude Opus 5, GPT-5.6 Sol, Grok 4.5, Kimi K3). Bring your own OpenRouter or provider key and heavy frontier coding bills your provider directly instead of drawing credits.
Can I drive it from my IDE or from another agent?
Yes. plori's MCP server lets any MCP client create an agent, hand it a task, poll the result, and answer its questions. Your IDE assistant can delegate a long refactor to a plori agent and pick up the result when it lands.
Is my code isolated?
Each agent runs on its own isolated machine with its own disk; nothing is shared across accounts. Deleting an agent permanently erases its disk.