Indigo loom · typed agents Framework · cloud · HITL

TypeScript agents · workflows · HITL

Knit the agent. Deploy the server.

Knitrel is a core AI/ML product: a TypeScript framework and cloud for long-running agents. You declare agents, tools, memory, and workflows in code, then deploy a typed server with traces, scorers, and human-in-the-loop pauses that can last days.

  • Agents typed objects
  • Workflows suspend and resume
  • HITL pauses as first-class steps
Claude 3.7 SonnetGPT-4oGemini 2.5Llama 3 70BBedrockAnthropicOpenAI DirectGroqLangGraph Claude 3.7 SonnetGPT-4oGemini 2.5Llama 3 70BBedrockAnthropicOpenAI DirectGroqLangGraph

AI features

Where the long-running agent actually lives

Knitrel is in the business of building a core AI/ML product for agents. Named jobs: typed agents, workflows with human-in-the-loop, memory, traces, and scorers. This is an agent framework and cloud — not a chat widget.

AI feature · Typed agents

Instructions, models, tools, and runtime in one object

For product engineers who write agents the way they write routes

Declare the agent in TypeScript. The typed object is the product. Run locally, then publish a durable server with a versioned URL, auth, and retries.

npx create-knitrel@latest → POST /v1/agents/{id}/runs

AI feature · Workflows and HITL

Suspend, resume, fan-out. Human pause is a step.

For teams whose agents wait on a person, not only a model

Long-running agents that can pause for days. Human-in-the-loop is a first-class step. Resume from the same run id.

run → HITL pause → POST /v1/runs/{id}/resume

AI feature · Memory, traces, scorers

Thread memory, workspace files, and evals next to the code

For operators who need provenance on every write

Thread memory plus workspace files. Traces, metrics, and scorers built in. Experiments live next to the code — not in a side spreadsheet.

GET /v1/runs/{id} → traces + scorers

AI feature · Model wiring

Claude, GPT-4o, Gemini 2.5, Llama 3 70B

For platform teams pointing a typed agent at the families they already call

Planned model families: Claude 3.7 Sonnet, GPT-4o, Gemini 2.5, Llama 3 70B. Access: Bedrock, Anthropic, OpenAI Direct, Groq. Agent frameworks: custom/homebrew and LangGraph. Knitrel is not fine-tuning those models.

Typed agent → selected model → Bedrock / Anthropic / OpenAI / Groq

AI development tools / APIs. Open source MIT framework plus hosted cloud. Bots, APIs, machine learning. MVP stage — no invented customers, funding, or uptime claims on this page.

Method

How Knitrel ships

Write the object. Run it locally. Publish a durable server.

01

Declare the knit.

Agents, tools, memory, and workflows in TypeScript. The types are the contract.

02

Run the loom.

Local studio first. Human-in-the-loop pauses are steps, not afterthoughts.

03

Publish the server.

A versioned URL with traces, scorers, and pauses that can last days.

Stack

What you actually buy

A TypeScript agent runtime. Framework free. Cloud by run-hour.

Agents

Instructions, models, tools, and runtime in one typed object.

Workflows

Suspend, resume, fan-out. Human-in-the-loop is a first-class step.

Memory

Thread memory plus workspace files. Provenance on every write.

Observe

Traces, metrics, and scorers built in. Experiments live next to the code.

Start

A typed agent this afternoon.

Write team@knitrel.xyz. You get a test token, a sample agent, and a 30-minute walkthrough on the workflow you want to run.

See pricing