SMELT

GIMZWARE / AI workflow builder

Visual multi-agent workflows

Multi-agent AI. No framework. No lock-in.

Design agent teams on a canvas, give them tools, knowledge and memory, and watch every run as it happens. Managed AI on our keys or yours. Build it, run it, own it.

  • Agent teams, orchestrated. Describe a goal and Auto Agent decomposes it into typed sub-tasks and wires the dependencies. Your own agents delegate to each other with agents as tools.
  • Managed AI, zero keys. Frontier and open model families, ready on our keys, or bring your own.
  • Rust-fast, Python-smart. Rust routes the hot path, Python runs the agents. Built for throughput, not Node.js.
Start freeView pricingRead the docs
PythonRustNATSPostgres

Free forever. No credit card required.

AI Models
Multiple
Integrations
Multiple
Triggers
Multiple
Engine
Rust + Python
Free tier
No card required

/ how it works

Describe the goal. The agents build the plan.

One sentence in. A typed sub-task plan out, with dependencies wired and the right agent on each step.

/ the old way

9 nodes, hand-wired

Every step is a box to configure. Search, parse, loop, filter, rank, compose, send. Each one a form to fill and a field to fat-finger.

  1. 01

    HTTP request

    search query, headers, retry

  2. 02

    Parse HTML

    CSS selectors, fallback

  3. 03

    For each result

    loop, batching, timeouts

  4. 04

    HTTP request

    supplier page, headers

  5. 05

    Function

    regex price + rating

  6. 06

    Filter

    expression: price < 15

  7. 07

    Function

    rank, slice top 3

  8. 08

    Set

    compose email body

  9. 09

    Send email

    per supplier, retry

/ with auto agent

1 block, 1 sentence

One block, one sentence. The orchestrator decomposes the goal into typed sub-tasks, picks which agent runs each, and wires the dependencies.

Auto Agent

research + reasoning, max 5 sub-tasks

Goal

Find the top three coffee suppliers in London under 15 GBP/kg and draft a short outreach email to each.

/ generated plan

  1. sub-1research
    no deps

    Find London coffee suppliers under 15 GBP/kg with public reviews

  2. sub-2research
    after sub-1

    Look up review aggregate + wholesale terms for each candidate

  3. sub-3reasoning
    after sub-2

    Rank candidates, pick the top three under the price ceiling

  4. sub-4reasoning
    after sub-3

    Draft three personalised outreach emails citing each supplier's specifics

Gmail / Send email

takes the drafted text, sends one email per supplier

Auto Agent produces the drafted text via research + reasoning sub-tasks. Side effects (sending email, creating events, writing files) stay in explicit, user-wired integration blocks downstream. An Approval Gate can sit between them if you want a human pre-flight.

Live plan rendering

The plan appears on canvas as the orchestrator emits it.

Dependency-aware plan

The plan records which sub-tasks depend on which, so independent branches are dispatched together rather than in a fixed order.

Inspectable plan output

Every plan is structured JSON with the orchestrator's reasoning attached.

/ delegation

Pro

Agents that work as a team.

Most builders stop at one agent per step. SMELT agents hand work to your other agents in the middle of a run, on a canvas you own, and each block pins the exact agent version it runs.

Give an agent your other agents as tools. When it delegates, the Trace shows the hand-off live: which sub-agent ran, how long it took and what it cost in credits. Deep chains stay visible, however deep the delegation runs.

  1. External interop

    The open A2A protocol to agents beyond SMELT

    Included

  2. Sub-agent delegation

    Agents call your other agents inside a run

    Included

  3. Integrations

    OAuth connectors plus your own MCP servers

    Included

  4. Memory

    Agents remember across runs, on stores built in

    Included

  5. Retrieval and research

    Knowledge bases with grounded, cited retrieval

    Included

  6. Models and tools

    Frontier and open model families with typed tool calls

    Included

Sub-agent delegation happens inside one run. To hand work to agents outside SMELT, use the Custom A2A connector.

/ the trace, live

one run, two agents

A supervisor delegates mid-run. The hand-off, the sub-agent and the credits appear in the Trace as they happen, and the replay shows the same tree afterwards.

  1. StartedSupport triage
    agent
  2. DelegatedRefund policy checker
    sub-agent
  3. Sub-agentRefund policy checker
    4.9s - 70 credits
  4. CompletedSupport triage
    12.0s

Credits roll up per sub-agent call. The supervisor node on the canvas shows who it delegated to for the whole run.

What powers Studio

Proven infrastructure, managed for you

We run the message bus, memory, routing and observability, so you get the reliability without the operations.

  • AI models
    Claude, GPT Series, Mistral, Llama
    Use our managed AI or bring your own OpenAI or Anthropic key
  • Workflow engine
    Python + Rust
    Python for agents, Rust for routing
  • Agent contracts
    Pydantic v2 (strict mode)
    Every input and output validated, no silent failures
  • Message bus
    NATS JetStream
    Durable delivery with replay for debugging
  • Memory
    Redis + Postgres + pgvector
    Working context, run history and semantic search
  • Integrations
    OAuth and API key
    Connect an account once and reuse it across workflows
  • Execution
    SSE real-time streaming
    Block-by-block streaming with live credits and duration
  • Observability
    OpenTelemetry
    Every agent call traced: model, latency, credits
  • Infrastructure
    Azure (managed)
    Managed containers, zero-downtime revision deploys

Connections

Connect and compose, in seconds

OAuth, REST and webhooks are built in. Pro adds your own MCP tool servers, external A2A agents, and agents that call each other. No credential plumbing.

OAuth connectors

Google / Microsoft / Slack / GitHub / Telegram, and growing

Built-in

One-click sign-in with the services your users already have.

REST APIs

HTTP with SSRF protection

Built-in

Call any external API from your workflows with built-in URL validation.

Webhooks

Trigger workflows externally

Built-in

Start a workflow from any system via HTTP webhook and receive results via callback.

Custom MCP

Bring your own tools

Pro

Point Studio at your private MCP servers to give agents access to internal tools.

Custom A2A

Connect external agents

Pro

Delegate to any A2A-compatible agent as a first-class collaborator inside your workflow.

Agents as tools

Agents call your agents

Pro

Compose specialists: let one agent invoke your other agents as tools within a single run.

Pro also brings agent memory, premium models and AI content safety. See what Pro includes

Pricing

Start free. Scale when you do.

No card required. Upgrade or cancel any time.

A run draws credits as it works, so both limits below apply and whichever you reach first pauses new runs until the month resets. Top up credits at any time, or connect your own model key to stop the credit meter entirely.

Prices are in pounds sterling and include any VAT we are required to charge, so the figure shown is the amount you pay.

Free

£0forever

  • Visual workflow builder
  • Starter templates
  • Execution monitoring (SSE)
  • Workflows3
  • Executions / month100
  • Managed AI credits / month75K
  • Custom agents2
  • Model accessStandard
  • Knowledge base (RAG)Text and web, standard embeddings
  • BYOK (bring your own key)
  • Integration connectors
  • Approval gates
  • SupportCommunity
  • Custom MCP and Custom A2A connectors: external tools and agents
  • Content Safety block and agent tool
  • Sub-agent delegation: agents call your other agents in a run
  • Agent memory: agents remember across runs
  • Team workspaces: share projects with colleagues
  • Builder seatsJust you
Start free
Best Value

Pro

£19.99/month

  • Visual workflow builder
  • Starter templates
  • Execution monitoring (SSE)
  • Workflows15
  • Executions / month10,000
  • Managed AI credits / month1M
  • Custom agents20
  • Model accessAll, incl. premium
  • Knowledge base (RAG)URL and file upload, premium embeddings
  • BYOK (bring your own key)
  • Integration connectors
  • Approval gates
  • SupportEmail
  • Custom MCP and Custom A2A connectors: external tools and agents
  • Content Safety block and agent tool
  • Sub-agent delegation: agents call your other agents in a run
  • Agent memory: agents remember across runs
  • Team workspaces: share projects with colleagues
  • Builder seatsJust you
Upgrade to Pro

Business

£59.99/month

  • Visual workflow builder
  • Starter templates
  • Execution monitoring (SSE)
  • Workflows50
  • Executions / month50,000
  • Managed AI credits / month4M
  • Custom agents100
  • Model accessAll, incl. premium
  • Knowledge base (RAG)URL and file upload, premium embeddings
  • BYOK (bring your own key)
  • Integration connectors
  • Approval gates
  • SupportPriority
  • Custom MCP and Custom A2A connectors: external tools and agents
  • Content Safety block and agent tool
  • Sub-agent delegation: agents call your other agents in a run
  • Agent memory: agents remember across runs
  • Team workspaces: share projects with colleagues
  • Builder seats3 included, then £15/month each
Go Business

Build your first workflow in minutes.

No card required.

Start free