AgentForge
AI Agent-as-a-Service. Build custom support agents, internal ops agents, and sales assistants that call your tools, remember context, and run behind a versioned API — without standing up your own model-serving or orchestration stack.
Everything an agent needs, none of the glue code
Tool calling
Register any REST endpoint, internal API, or database as a typed tool. AgentForge handles the function-calling loop, retries, and argument validation against your schema.
Model routing
Pick a Bedrock model per agent, or let AgentForge route by task complexity and cost budget. Swap models without changing your integration code.
Session memory
Short-term session memory out of the box, backed by DynamoDB, with configurable retention. Bring your own vector store for long-term recall.
Guardrails
Attach Bedrock Guardrails policies per agent — PII redaction, topic denylists, jailbreak detection — and get a webhook event whenever one trips.
Versioned deploys
Every change creates a new agent version behind the same endpoint. Roll out gradually, A/B two prompt versions, or roll back instantly.
Run tracing
Every invocation, tool call, and model response is logged to S3 with a trace ID. Replay a failed run locally against the same inputs.
From API call to answer, in five steps
Request hits API Gateway
Your app calls agent.invoke(). API Gateway authenticates the request against your API key and applies per-key rate limits.
Lambda loads agent config
An orchestrator Lambda resolves the agent's current deployed version — its model, tools, memory settings, and guardrail policy.
Bedrock generates a response
The configured model runs on Amazon Bedrock. If the model requests a tool call, Step Functions coordinates the call-and-resume loop.
State is written to DynamoDB
Session memory and agent state persist to DynamoDB so the next turn in the conversation has context, even across cold starts.
Trace and response return
The final answer streams back to your app, while the full trace — prompts, tool calls, latency — lands in S3 and your dashboard.
Where teams put AgentForge to work
Customer support agents
Answer order status, refund, and account questions by calling your helpdesk and order APIs directly, with a defined handoff-to-human path.
Ops and on-call assistants
Query internal dashboards, summarize incident channels, and open tickets — scoped to internal tools with audit-logged actions.
Sales and success assistants
Draft follow-ups from CRM context, qualify inbound leads against your ICP, and surface account health before renewal calls.
Deploy an agent in one API call
The same primitives — create, invoke, deploy — work whether you're calling the REST API directly or using the TypeScript/Python SDK.
# create an agent from a config file
curl -X POST https://api.qxentrixai.com/v1/agents \
-H "Authorization: Bearer $QX_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"name": "ops-agent-v1",
"model": "bedrock:anthropic.claude",
"tools": ["pagerduty", "internal_wiki"],
"guardrails": "internal-tools-v1"
}'
# promote a version to the live endpoint
curl -X POST https://api.qxentrixai.com/v1/agents/ops-agent/deploy \
-H "Authorization: Bearer $QX_API_KEY" \
-d '{ "version": 3 }'Start free, scale with usage
Every tier includes the full agent runtime. Higher tiers add concurrency, SLAs, and dedicated support. See the full pricing page for a monthly/annual breakdown and FAQ.
For teams in production
- 10 active agents
- 100,000 invocations / mo
- Guardrails + webhooks
- Email + Slack support
For scaled deployments
- Unlimited agents
- Volume-based pricing
- VPC / private endpoint
- Dedicated support + SLA