You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The CLI names its agent CloudThinker Agent everywhere: agent starts CloudThinker Agent in this folder, and chat asks CloudThinker Agent in the cloud.
cloudthinker incident ls lists the workspace's incidents, and cloudthinker incident status <INCIDENT_ID> --wait polls until the incident is resolved, dismissed, or not found, so an on-call script or CI job can wait on it. It exits 4 on timeout; add --json for a script to read.
cloudthinker recommendations ls lists Optimize recommendations with their monthly savings, filtered by --status, for cost reports.
cloudthinker chat -p accepts --agent for an authorized custom-agent UUID or alias and keeps model selection separate.
Keep the generated CLI session and capability contract aligned with custom-agent selection and the specialist flag.
cloudthinker incident status --help now separates the exit codes: 0 when the incident closes (including the NOT_FOUND status), 1 when no incident has that ID.
The bundled Cyber discovery runtime accepts browser_resource_origins in a target scope, matching the backend.
The CLI's API contract no longer carries the incident declaration actor, which the CLI never reads.
Install cloudthinker-cli 0.8.0-dev.4
Install prebuilt binaries via shell script
curl --proto '=https' --tlsv1.2 -LsSf https://github.com/cloudthinker-ai/cloudthinker-cli/releases/download/v0.8.0-dev.4/cloudthinker-cli-installer.sh | sh
The agent opens fullscreen with a status line for workspace, approval mode, credits, and Connections; Ctrl+O opens a searchable transcript, and a cloud write asks in one approval modal.
A bare cloudthinker --tui-mode regular or --tui-mode fullscreen picks the agent screen without the agent subcommand.
cloudthinker agent --help shows CloudThinker's own help, and the top-level help calls it the CloudThinker agent. Flags and commands the agent does not support now stop with the next step instead of running.
In the agent, /share copies the conversation link, /login and /logout sign in and out of CloudThinker, /changelog shows the agent's releases, and /bug opens a GitHub issue.
The generated API client picks up the code review layer_count field.
cloudthinker --help lists what a person runs. Commands only an AI agent drives (cloud, and the Cyber run, probe, discovery, and memory steps) still work but are hidden; cloudthinker --skill <module> documents them.
Every command and option now has help text, Cyber help is written for users instead of describing internals, and every --json reads the same.
chat is described as the cloud agent and agent as the agent in this folder, so the two headless paths (chat -p and agent -p) are easy to tell apart.
login --help shows one sign-in option, --device-auth, for a short code; --no-browser still works but is no longer listed.
worker service commands default --workdir to the current folder and --outpost to the only outpost set up on this machine.
cloudthinker agent --help prints the agent's help at once, without downloading the agent.
Install cloudthinker-cli 0.8.0-dev.3
Install prebuilt binaries via shell script
curl --proto '=https' --tlsv1.2 -LsSf https://github.com/cloudthinker-ai/cloudthinker-cli/releases/download/v0.8.0-dev.3/cloudthinker-cli-installer.sh | sh
Add direct cloud commands for independent agents: create sessions, discover Connections, load their guides and SDK schemas, execute scoped scripts, and recover task output or approved writes. Print release-matched guidance with cloudthinker --skill cloud.
Render completed Mermaid flowcharts as themed terminal diagrams with authored node border colors; keep interiors and text in terminal theme colors. Preserve source while streaming and when a preview cannot fit or uses unsupported syntax.
Install cloudthinker-cli 0.8.0-dev.2
Install prebuilt binaries via shell script
curl --proto '=https' --tlsv1.2 -LsSf https://github.com/cloudthinker-ai/cloudthinker-cli/releases/download/v0.8.0-dev.2/cloudthinker-cli-installer.sh | sh
Add direct cloud commands for independent agents: create sessions, discover Connections, load their guides and SDK schemas, execute scoped scripts, and recover task output or approved writes. Print release-matched guidance with cloudthinker --skill cloud.
Render completed Mermaid flowcharts as themed terminal diagrams with authored node border colors; keep interiors and text in terminal theme colors. Preserve source while streaming and when a preview cannot fit or uses unsupported syntax.
Make authored Mermaid border colors opt-in through markdown.mermaidBorderColors, defaulting to off.
Install cloudthinker-cli 0.8.0-dev.1
Install prebuilt binaries via shell script
curl --proto '=https' --tlsv1.2 -LsSf https://github.com/cloudthinker-ai/cloudthinker-cli/releases/download/v0.8.0-dev.1/cloudthinker-cli-installer.sh | sh
Render completed Mermaid flowcharts as themed terminal diagrams with authored node border colors; keep interiors and text in terminal theme colors. Preserve source while streaming and when a preview cannot fit or uses unsupported syntax.
Install cloudthinker-cli 0.7.7
Install prebuilt binaries via shell script
curl --proto '=https' --tlsv1.2 -LsSf https://github.com/cloudthinker-ai/cloudthinker-cli/releases/download/v0.7.7/cloudthinker-cli-installer.sh | sh
Start with /skill:tour: a bundled, agent-led guide to Local and Cloud that helps you choose a first task, including with Cloud off.
Keep automatic skill metadata within an 8 KiB context budget. Omitted skills remain searchable and callable through /skill:name; budgeting works automatically.
Release Notes
Worker glob_read reads the head of a large file instead of failing the whole scan, so a big SKILL.md no longer hides the Worker folder's Skills.
On 64-bit Arm Linux, the credential lock and the worker skill bundle cache lock now refuse a symlink instead of following it, as they already did on x86_64 Linux and macOS.
Reduce worker memory use: file reads, operation dispatch, receipt journaling, and artifact uploads no longer copy operation payloads or file contents; journal bytes and JSON output are unchanged.
Worker file search, grep, and skill bundle verification do less repeated work on large folders; their results are unchanged.
A server error or rate limit during a token refresh is retried, honouring Retry-After, instead of logging you out; device-code login keeps polling through a server error.
A refresh keeps your active workspace; it no longer switches to the workspace you last used with --workspace.
An older cloudthinker never overwrites a credentials file written by a newer one, and an unreadable credentials file is moved aside before a login replaces it.
A 403 shows the server's reason and asks for a workspace admin instead of sending you to cloudthinker login.
Every login hint names the host you used, so cloudthinker --url https://dev.cloudthinker.io ... tells you to run cloudthinker login --url https://dev.cloudthinker.io.
New cloudthinker auth status lists your stored workspace logins, and cloudthinker auth switch <workspace> makes one active.
logout says why a server revoke failed, says when nothing was stored, names the stored workspaces when none is active, and warns while CLOUDTHINKER_TOKEN is still set.
cloudthinker auth token finishes a token refresh it started before it stops on Ctrl-C, and the agent uses the same cloudthinker binary for its token.
Commands no longer read the OS keyring once the credentials file holds the host's login.
Worker credential errors say which check failed, and a missing outpost credential asks you to register the outpost again.
worker start keeps running through short CloudThinker outages: it retries server errors and rate limits, reconnects after a laptop sleep or network change, and no longer cancels running work on one bad gateway.
A finished operation result is retried until its lease ends, so it is no longer marked unknown after five seconds of network trouble.
A machine clock that runs ahead or behind no longer fails outpost work.
The worker log names each assignment and operation with its duration and error code, and server errors show their code and request ID.
The first Ctrl-C drains running operations and says how many are left; a second Ctrl-C cancels them.
Shell output that was printed before a detached background process kept the output open is now returned instead of lost.
The skill bundle cache frees bundles unused for 25 hours, so new skills keep installing after 64 versions.
worker service status shows the last exit code and where the logs are, the macOS service now writes a log file, and service manager errors show their reason.
A replaced served folder stops the worker with a clear message; worker start --reset-folder serves the new folder.
worker service install updates a service that an earlier release installed with the same settings and reports updated, instead of asking you to uninstall it first.
login over SSH or on Linux without a display now shows a short code to enter in any browser, instead of waiting 5 minutes for a browser callback that cannot arrive. Set CLOUDTHINKER_LOGIN=browser to keep the browser callback.
login --url <address> now remembers that address, so later commands and login hints reach the same CloudThinker without --url. Logging out of that address returns the default to production.
chat -p now adds piped input to the prompt, for example kubectl logs pod | cloudthinker chat -p "why does this crash?", and chat -p - reads the whole prompt from stdin.
chat -p now shows a spinner with the run status and elapsed time on a terminal. Ctrl-C stops the wait and prints the command that resumes it.
chat --json now includes message and failure_kind, so a script can see why a run failed.
review --fail-on <severity> and review watch --fail-on <severity> exit 6 when a finding reaches that severity, so a CI job can gate on a review.
New cloudthinker completion <shell> prints a completion script for bash, zsh, fish, PowerShell, or elvish.
chat, review, login, logout, and whoami now end with one line when a newer release exists.
whoami --json now includes the account email and the workspace name, and chat ls says so when there are no runs.
--help now opens with a short get-started list, and a failed update names the reinstall command.
Run local Cyber discovery through the shared background command engine; reserve workflows for agent investigation and verification. Discovery cancellation also stops collectors created during cleanup.
Start local pentests from the interactive agent with automatic setup checks, configuration repair, visible progress, and readable results.
Add internal Cyber commands for targets, runs, scoped probes, coverage, identities, findings, and evidence. The backend owns the plan and final result.
Keep coverage readable after a run finishes while preventing further plan or observation changes.
Add the local development launcher, separating backend and target credentials and keeping token refresh within the agent deadline.
Omit cloud-only memory paths from local run briefs and report missing coverage reasons before authentication or network calls.
Keep a private per-run local workspace for resumable evidence drafts and workflow journals; repeated binds reuse it, while canonical App memory remains the cross-run source.
Match saved target paths using backend trailing-slash normalization without changing query values.
cyber memory pull <app-id> --output <dir> saves the App's canonical findings and surface to a local directory. --include-context downloads non-credential App documents with explicit opt-in; context files are bounded and their signed URLs are never written to the snapshot.
Identify the local CLI build as 0.7.6-dev.1, including background command execution and corrected Cyber workflow guidance.
Local Cyber workflows run the shared discovery collectors on your machine, preserve discovery limitations, and require a valid report before successful completion.
Local source-agent runs can use the launcher-selected auth binary for token resolution while the production agent keeps the wrapper's own native binary.
Keep the CLI API client aligned with password authentication and review job commit counts after the develop merge.
cloudthinker-local defaults workflow concurrency to 4; set CLOUDTHINKER_WORKFLOW_MAX_CONCURRENCY=8 to use the shipped default. Values must be from 1 to 8.
Install cloudthinker-cli 0.7.6
Install prebuilt binaries via shell script
curl --proto '=https' --tlsv1.2 -LsSf https://github.com/cloudthinker-ai/cloudthinker-cli/releases/download/v0.7.6/cloudthinker-cli-installer.sh | sh
Review the local checkout with a bundled read-only Pi agent using CloudThinker inference; keep the prompt, transcript, findings, and rendered result local.
Confine the review agent's file tools to existing, non-ignored paths within the checkout and hide Git metadata and ignored paths from tool results.
Enforce worker transport retry deadlines even when fetch completion and the deadline become ready before the runtime is polled.
Install cloudthinker-cli 0.7.3
Install prebuilt binaries via shell script
curl --proto '=https' --tlsv1.2 -LsSf https://github.com/cloudthinker-ai/cloudthinker-cli/releases/download/v0.7.3/cloudthinker-cli-installer.sh | sh
