Skip to content
New · AI Config Releases

Change how agents behave.
Keep control.

Release prompts, models, tools, and skills to the right users without redeploying your application or sending model traffic through FlagDash.

Start releasing safely Read the guide
support-agent / production● live
Baseline90%
gpt-4o-mini · support-v2
guarded release
Candidate10%
claude-sonnet · support-v3
latency ✓quality ✓cost ✓
application calls the provider directly

Two products, one clear boundary

AI Config is the artifact.
AI Releases is the decision.

AI Config

Manage prompts, agent instructions, rules, and Agent Skills as readable files. Organize them per environment and load them through API or MCP.

AI Config Releases New

Choose a configuration for each user, roll it out gradually, observe signals, and return to the baseline instantly when it is not safe.

How it works

A safer loop for every AI change.

Your application still calls OpenAI, Anthropic, or another provider. FlagDash only decides which configuration the application should use.

01

Define

Save model, prompt, tools, parameters, and skills as a named configuration.

02

Target

Use user, plan, region, segment, or percentage rules to choose a variant.

03

Observe

Send cost, latency, quality, safety, or business signals back to the release.

04

Act

Advance, pause, or roll back without a deploy when the evidence changes.

For the whole team

One release surface.
Developers, product, and the business.

release graph / live

support-agent

healthy
Build

v3

prompt + tools

Target

10%

candidate cohort

Signal

Pass

latency · cost

Developers

Keep provider calls in your application, but move targeting and AI-version branching out of your code.

Product teams

Test prompts, models, and tool policies against outcomes—not intuition or a lab-only score.

Companies

Limit exposure, control spend, preserve an audit trail, and recover quickly when behavior changes.

The boundary matters

FlagDash selects.
Your app requests.

No model keys in FlagDash. No user prompts routed through another service. Your existing OpenAI or Anthropic SDK keeps making the request with the configuration FlagDash returned.

See the integration guide
config = flagdash.ai_config("support-agent")
response = openai.chat(config)
✓ prompt v3 selected
✓ candidate rollout: 10%
✓ provider call stays in your app