Developer Tools

camelRun

by camelAI

Overview

camelRun is camelAI's hosted runtime for AI agents. Connect your camelRun account to create agents from a plain-language description, give them built-in tools such as web search, run them and read their replies, and deploy agent.yaml manifests that keep an agent's configuration as a reusable, versioned definition. Agents are durable: each keeps its history until you delete it, can wake itself on a schedule, and can stop to ask you a question or for an approval, which you answer from the conversation. It is for developers and teams who build agents for their products or their own work and want to try, adjust and manage them without leaving the conversation. The plugin acts only on the camelRun account you sign in with, and only on its agents, definitions and schedules. Runs use your account's credit. It does not manage billing, API tokens or provider keys; do those in the camelRun console. Tool servers that need secrets are best deployed with the camelrun CLI, which reads them from its environment, so credentials never pass through the conversation. Deleting an agent or definition cannot be undone, so you are asked to confirm first.

Tools

abort_agent

ChatGPT

Stop an agent's running turn; the turn's unfinished work is cancelled. The agent and its history stay.

add_schedule

ChatGPT

Wake an agent later with a message: once (inSeconds or at), or repeatedly (everySeconds, at least 60). Each wake-up is a run that uses the account's model credit.

agent_history

ChatGPT

An agent's latest messages, in whole turns: text, tool calls and tool results. Changes nothing.

answer_input

ChatGPT

Answer an input and wait for the run it resumes. An approval takes true or false; a question the chosen label or your own words (several questions: {"<question>": "<answer>"}); a form its fields; "decline" declines. Only answer with what the person decided: an answer cannot be taken back, and an approval lets the agent go ahead with what it asked to do.

configure_agent

ChatGPT

Change one agent's model, system prompt, prompt addition or thinking level between its runs, replacing the current values; its history is kept.

create_agent

ChatGPT

Make the agent for a key: from a definition, or with its own model and system prompt. The same key is the same agent, so calling it again for a key that exists updates that agent to the configuration given, keeping its history. It lives until deleted.

definition_agents

ChatGPT

The agents made from a definition, and the revision each has. Changes nothing.

delete_agent

ChatGPT

Delete an agent: it stops at once, and its history and files are purged; this cannot be undone, so confirm with the person first. Its key makes a fresh agent next time.

delete_definition

ChatGPT

Delete a definition and its revisions; this cannot be undone, so confirm with the person first. Agents made from it keep running as they are.

delete_schedule

ChatGPT

Cancel a scheduled wake-up; this cannot be undone.

deploy

ChatGPT

Deploy agent manifests: upsert each definition by its key (a new revision only when it changed, replacing the current one) and make or update the keyed agents it lists. Use dryRun first to show what would change. Give manifest, the YAML itself (as an agent.yaml holds it). This server reads no files and no environment variables: put the system prompt in systemPrompt and any OpenAPI spec in spec. Never ask a person to paste credentials into the conversation: a manifest whose tool servers need secrets is deployed with the camelrun CLI, which reads them from its environment.

fork_agent

ChatGPT

Make a new agent from an existing one: the same configuration, a copy of its conversation and a copy of its files, each its own from then on, so the copy can try another direction while the original stays as it was. By default the copy ends with the last turn that finished; atMessage ends it earlier. With a key it lives until deleted, and calling again with the same key returns the same fork.

get_agent

ChatGPT

An agent's configuration, the definition revision it has, and every source of its tools with what each offers. Changes nothing.

get_definition

ChatGPT

A definition in full: model, system prompt, built-ins and tool sources (never their credentials). Changes nothing.

get_run

ChatGPT

A run's status and result, waiting up to wait seconds for it to end. Changes nothing.

list_agents

ChatGPT

The account's agents: key, id, name and model. loaded is whether the agent is in a runtime's memory now, idle or not; it says nothing about whether a run is going (run_agent and get_run do). toolsConnected is whether an application serving its attached tools is connected. Changes nothing.

list_definitions

ChatGPT

The account's definitions (reusable agent configurations): id, key, revision, model and built-in tools. Changes nothing.

list_inputs

ChatGPT

Questions, approvals and forms waiting on a person: one agent's, or every agent's. Changes nothing.

list_models

ChatGPT

The models agents can use, with context window, reasoning support, price and toolCallStreaming (true: tool-call arguments stream as written, so a page can render a section while it is written; false: they arrive in one piece; "unknown": not measured): pass a model's id as model to create_agent, configure_agent or a manifest. Changes nothing.

list_schedules

ChatGPT

An agent's scheduled wake-ups. Changes nothing.

read_docs

ChatGPT

camelRun's own documentation as Markdown: with no path, the index (llms.txt); else a page such as guides/definitions.md, guides/tools.md or reference/cli.md. Changes nothing.

run_agent

ChatGPT

Send an agent a message and wait for its run: the reply, the input it waits on, or why it failed. A run still going after wait seconds comes back as running; call get_run. The agent acts with the tools it was given (such as web search and its MCP servers), and the run uses the account's model credit.