See how your agents, skills, and harness
works and what they're focused on.
Agents run on skills and prompts, then work mostly out of sight. Vaizer makes the whole loop legible: read a skill before you run it, analyze and version the prompts behind it, control what each environment serves, and watch every session from one HUD.
One place to manage your agents
Break down any skill into a workflow
Skills are Markdown and tools; their real behavior is buried in prose. Vaizer renders any skill as a node graph, trigger to output, so you can read what it does and where it branches before you trust it with your machine. Browse the catalog, or paste any public skill and break it down.
Explore skillsWrite, version, and analyze your prompts
A prompt library with history, plus instant analysis: the weak points and how to fix them, the decisions the prompt quietly hands to the model, what a run costs on each model, and which model should run it, tuned to whether you want the cheapest fit or the highest quality.
Open the workbenchFeature-flag your prompt config
The LaunchDarkly workflow, pointed at prompts. Toggle what each environment serves, promote a prompt from development to production without a deploy, and cache or invalidate served prompts on your schedule.
See prompt configRun every session from one command center
Long overnight loops and quick chats, side by side, with progress and spend per session. Self-running agents stay quiet while they work; the attention queue surfaces the few that actually need a human. For a single long run, the Watch view replays it as a journey toward the goal.
Open the HUDBuilt on one idea: You can visualize anything.
Read before you run
Every skill becomes a graph you can click through, and every prompt gets a review: what it will do, what it leaves to chance, and what it costs.
Change without deploying
Prompts are config, not code. Version them, flag them per environment, and invalidate caches when you say so.
Watch the momentum
The HUD shows every session and who needs you; Watch replays one long run as a path of milestones and attempts.
Stop guessing what your agents are doing.
Analyze the prompt, flag the config, and keep every session in view.