P
Penseur
Get early access
⚡NewDesign, deploy & orchestrate AI agents – visually.

Build production agents without stitching APIs.

Penseur is the control plane for AI agents. Design flows with Anvil, orchestrate backends with Forge, and deploy to your stack in minutes – no brittle glue code, no YAML sprawl.

Trusted by teams shipping support, research and ops agents.
● Audit‑ready logs
Agent canvaslive preview
●●●
Toolbox12 tools
HTTP requestGET
Postgres querySQL
Slack notifychat
Policies
  • • Limit spend to $0.02 / request
  • • Strip PII before logging
  • • Auto‑fallback on tool failure
Customer support agent
live
Entry
User message
Reason
Routing agent
Action
Knowledge search
Guardrail
Compliance filter
live trace · request #129008latency: 1.7s · cost: $0.004
> user: "Can you update the shipping address on my last order?"
✔ route → support_flow
↳ tools.http: GET /orders/last
↳ tools.vector_search: kb/support‑policies
↳ guardrail.compliance: ok
✔ response drafted
Export agents as

Everything you need to ship real agents

From prototypes to production: orchestrate tools, control behavior and keep observability built‑in.

SOC2 underway · GDPR‑ready
⧉

Anvil – Visual frontend

Design agent flows visually with drag‑and‑drop canvas. Define tools and route between flows effortlessly.

  • • Graph view for multi‑step reasoning
  • • Environment‑aware configuration
  • • Versioning & safe rollbacks
🛠

Forge – Backend power

Orchestrate HTTP, DBs, vector stores and custom functions with built‑in retries, timeouts and rate limits.

  • • Strongly typed inputs / outputs
  • • Tool usage analytics & limits
  • • Secrets managed via your vault
🛡

Guardrails & observability

Keep agents on‑policy with structured control, full traces and red‑teamable behavior logs.

  • • Policy checks per step or flow
  • • Trace viewer & replayable sessions
  • • Fine‑grained evaluation hooks

From idea to shipped agent in 3 steps

Design your agent once, then run it anywhere – API, SDK or your own infra.

1

Model your flow

Outline how your agent should behave: routing logic, tools it can call and which outputs matter.

  • • Choose entrypoints (chat, HTTP, cron)
  • • Define sub‑agents per responsibility
  • • Configure temperature, context & memory
2

Connect tools & data

Plug in your APIs, databases and knowledge bases. Add guardrails where it matters.

  • • HTTP, SQL, vector stores, queues
  • • Structured error handling & fallbacks
  • • Input/output validators
3

Deploy & iterate

Promote to production with one click, then observe traces, tune prompts and ship new versions safely.

  • • REST & WebSocket API
  • • TypeScript / Python SDKs
  • • Canary deploys and A/B tests
sdk/agent.tsready to paste
// TypeScript SDK
import { Penseur } from "@penseur/sdk";

const client = new Penseur({
  apiKey: process.env.PENSEUR_KEY!,
});

const supportAgent = client.agent("support_router");

const result = await supportAgent.run({
  userId: "user_123",
  input: "Can you update the shipping address on my last order?",
  context: {
    channel: "chat",
    locale: "en-US",
  },
});

console.log(result.message);
console.log(result.tools);
SDKs are open‑source & self‑host friendly.

Built for real workloads

Penseur powers agents across support, operations, research and internal tools.

Customer support agent

–40% handle time

Auto‑resolve tickets, summarize threads and handle order changes with guardrails on policy and refunds.

  • • Zendesk / Intercom integration
  • • Order & billing tools
  • • Escalation workflows

Ops automation agent

3x faster playbooks

Let agents watch metrics, trigger runbooks, create incidents and update dashboards automatically.

  • • PagerDuty / Slack integration
  • • Safe, audited tool execution
  • • Approval workflows

Research & analysis agent

hours → minutes

Multi‑step research agents that browse, summarize, compare and write reports your team can trust.

  • • Web, PDFs, internal docs
  • • Source‑grounded answers
  • • Citations & exportable outputs

Answers to common questions

Still unsure? Ask a Penseur engineer in onboarding — or start free and try it yourself.

Do I need to change my LLM provider?+
No. Penseur is model‑agnostic. Use OpenAI, Anthropic, Google, open‑source models, or your in‑house stack. You can mix models per agent or per step and swap them without changing your app code.
Can I deploy Penseur locally?+
Yes. While the control plane is a managed cloud service, a self‑hosted deployment environment for local infrastructure is available.
How do you handle security and privacy?+
We never store your raw prompts or data unless you explicitly enable logging. PII scrubbing, encryption at rest and in transit, and fine‑grained access controls are available out of the box. SOC2 is in progress.
What does "agent steps" mean for billing?+
A step is a single decision or tool call an agent makes: an LLM call, a tool invocation or a routing step. Your model costs stay with your provider — we only charge for orchestrating the agent.