Compose the agent
Define identity, prompt, model, response schema, tools, MCP servers, knowledge bases, connectors, and memories as one versioned unit.
Cloudflare-native agent workbench
Flaredin
Build, run, and improve production AI agents without turning your stack into glue code. Flaredin brings prompts, tools, MCP, knowledge, memory, traces, evals, approvals, and deployments into one operator-grade workspace.
Build
Agents, prompts, tools, MCP, KBs, skills
Observe
Traces, scores, evals, runs, files
Govern
Secrets, access, approvals, audit, billing
The platform
The agent control plane for teams shipping on Cloudflare.
Flaredin treats agents like software you can inspect, test, version, deploy, and repair. It is not a chatbot skin. It is the workspace around the agent loop.
Watch the run
Every turn records timing, tokens, tool calls, approvals, cost signals, and trace links so debugging does not depend on log archaeology.
Control the change
Ship versions into environments, gate risky tools, scope API keys, manage secrets, review audit events, and let Ember propose changes for approval.
Workbench
One agent, seven surfaces, no mystery state.
The current Flaredin app is organized around the lifecycle your team actually uses: chat with the agent, edit its identity, attach capabilities, tune runtime state, connect channels, govern access, and inspect runs.
agent workspace / support-ops / v4 draft
Identity
prompt + model
system prompt
model
temperature
response schema
voice
Capabilities
what it can do
HTTP tools
MCP
knowledge
skills
connectors
Runtime
state + triggers
memories
cron
webhook
email-in
Connect
external entrypoints
web widget
HTTP API
agent MCP
Slack
Telegram
Discord
WhatsApp
Governance
before production
secrets vault
agent access
approval rules
deploy slots
eval gate
research loop
audit log
Ember
The operator assistant that stages changes, not surprises.
Ember is Flaredin's meta-agent. Ask her why an agent is slow, where it is failing, or how to make a prompt safer. She reads operational context, writes investigations, and stages proposals you review before they land.
Prompt edit
Tighten escalation language after repeated approval failures in recent traces.
Eval dataset
Turn the agent's prompt goals into realistic input and expected-output cases.
Investigation note
Persist what Ember learned so the next debugging conversation starts ahead.
Observe and improve
Every run leaves enough evidence to fix the next one.
Traces show the full turn. Scores show human feedback. Evals pin examples to agent versions. Deployments pin versions to environments. Together, they give the team a loop: inspect, improve, test, ship.
DatasetRefund edge cases, 18 examples, substring matcher
ready
Runsupport-ops v4 against staging candidate
16 / 18 pass
DeployPin v4 to staging, compare against production v3
diff first
trace_8e1c / ops-triage / production
user turn0.00s
kb.query0.21s
model0.91s
tool approvalhold
scorepass
~/agents/support-opsflaredin
# create and test an agent
$ flaredin agent create --name Support --slug support
$ flaredin agent test ag_123 --message "Can I return this?"
ok trace created with tool calls and approvals
# ask Ember to inspect operations
$ flaredin ember chat "Why did support fail last night?"
ok 2 proposals staged for review
# ship when the version is ready
$ flaredin agent deploy ag_123 --env production
$ flaredin agent create --name Support --slug support
$ flaredin agent test ag_123 --message "Can I return this?"
ok trace created with tool calls and approvals
# ask Ember to inspect operations
$ flaredin ember chat "Why did support fail last night?"
ok 2 proposals staged for review
# ship when the version is ready
$ flaredin agent deploy ag_123 --env production
Developers
Use the dashboard for inspection. Use the API when it belongs in code.
Flaredin already exposes a TypeScript SDK and CLI around the REST API. The app is the workspace; the API is how teams wire agent creation, deployment, testing, and Ember review into their own workflows.
TypeScript SDK
Create agents, deploy versions, trigger tests, and chat with Ember from Node, browsers, and Bun.
CLI
Manage agents, configure org context, review Ember proposals, and request ranked fix recommendations.
Scoped access
Mint API keys for external callers, bind secrets to tools, and keep audit trails around operational changes.
Product map
Everything a production agent needs has a place.
Flaredin's workspace is organized into build, observe, and govern surfaces so teams can move from first prompt to production agent without another stack diagram.
Build
Agents
Chat
Prompts
Skills
Tools
MCP
Connectors
Knowledge
Memories
Secrets
Observe
Traces
Files
Scores
Evals
Approvals
Deployments
Webhooks
Govern
Team
Audit
Billing
Access
Approval rules
Research loop
Ember investigations
Make the agent visible, testable, and shippable.
Flaredin is the workbench around your Cloudflare-hosted agents: build the loop, connect the tools, watch the traces, evaluate versions, approve changes, and tell us what should feel faster next.