Agents and workflows

Agents that run your processes, in the systems you already use.

Spaceplace turns a process your team runs by hand into a workflow that agents carry out. They answer in your channels, act in your CRM and ERP, follow your rules and hand decisions to your people. Your team builds it, or our engineers build it with you.

What the buyer sees
RRiverside ResidencesSales chat

Hi! Do you still have two-bedroom flats in building 2? A high floor, if possible.

Sales agent

Yes, three are free in building 2, from the 9th floor up. No. 94 has the best view: 64 m², $118,000. Would you like to see it?

Yes. Thursday after 3 works for me.

Sales agent

Booked: Thursday at 15:00, building 2. I've sent you the address and your manager's name.

Can you do 10% off if I pay in full?

Sales agent

Discounts are decided by our head of sales. I've passed her your request.

Olena, head of sales, joined the chat
Olena · Head of sales

Hi! For full payment we can offer 5%. Shall I hold No. 94 for you until Thursday?

What happens behind it
spaceplaceAgentsWorkflowsEvaluation LabKnowledgeLatest published · v12
Lead qualificationRun 1,284RunningWaiting for a personTaken over by Olena
New messageTrigger · On Message
Wants to buy?Detect IntentBuy intentDiscount
Sales agentAI Chat · Auto (router)RepliedHanded to Olena
⇄unit_check_availabilityCall Skill✓ 3 free
⇄crm_create_leadCall Skill✓ Lead #2291
Reply to the buyerOutput
Agent reasoning
  • Message 1 · 14:02
  • IntentBuy · two-bedroom, building 2
  • SourcePrice list · building 2
  • Skillunit_check_availability → 3 free
  • GuardrailPassed
  • Message 2 · 14:05
  • Skillcrm_create_lead → Lead #2291
  • SynthesisViewing confirmed, Thu 15:00
  • Message 3 · 14:08
  • IntentDiscount request
  • EscalationKeyword «discount» → head of sales
  • Take overOlena · head of sales
Model & Cost · Claude Sonnet 5.5 · 1.1¢ this conversation
Capabilities

Everything an agent needs to do real work

The runtime, the builders, testing, models and control that every agent and workflow on Spaceplace shares. Each card says what we mean by it.

One turnAgent OS
Context
  1. System Preamble
  2. Custom Instructions
  3. Safety Preamble
  4. Knowledge Base (RAG)
  5. User Memory
  6. Tool Definitions
  7. Skill Descriptions
Then
  1. ModelClaude Sonnet 5.5 plans the reply
  2. Tool callunit_check_availability
  3. GuardrailPassed
  4. ReplySent to the buyer
  5. MemorySaved to the buyer's history
01 · Agent OS

Agent runtime

The engine that runs every agent. For each reply it gathers what the agent should know, lets the model use your tools step by step, checks the result against your guardrails and keeps what it learned for next time.

New agent · Sales assistant
BriefPlanSandboxEvaluatePublish
What should this agent do?

Recommended skills
unit_check_availabilitycrm_create_leadmortgage_calculator
✓ Passed the quality gatePublish
02 · Build

Agent Builder

Describe the job in plain words, and the builder drafts the agent: its instructions, the skills it needs and a first test, before it talks to anyone. Your team builds and releases; our engineers can build with you or advise.

Skill LabManual · Templates · Import
Add skill
  • MCP Tool
  • MCP Server
  • Function
  • Prompt
  • Web Search
  • External Agent (A2A)
Tool · Reversibility
unit_check_availabilityread_only
crm_create_leadreversible
issue_refundirreversible
Irreversible: waits for a person to confirm
03 · Skills

Skill Lab

A skill is one thing an agent can do: check a flat, create a lead, search the web, ask another agent. Skill Lab is where skills are made and connected, and every tool says whether its action can be undone.

Support agent v4 · your casesJudges: etalon · faithfulness · safety
  • Is flat 94 still free?Agent answering…Judge scoring…Pass
  • Can I pay in instalments?Agent answering…Judge scoring…Pass
  • Do you give 10% off?Agent answering…Judge scoring…FailNEW FAIL
  • Which floors have a view?Agent answering…Judge scoring…Pass
Pass rate 94%Your quality bar
04 · Testing

Evaluation Lab for agents

Before an agent goes live, it answers your real cases and judge models score every answer: is it right, is it grounded in your sources, is it safe. A new version that does worse does not ship.

Routing mode
Auto (router)Pinned
TaskCheck a supplier contract against our policyWhere is the refund letter template?
  1. FrontierClaude Opus 5.5
  2. PremiumClaude Sonnet 5.5GPT-6.1
  3. StandardClaude Haiku 5.5
  4. Nanoqwen2.5:7bOn your servers
05 · Models

Multi-model by design

No agent is tied to one model. The router sends each task to the tier that fits it, from a small model on your own servers to the strongest one, or you pin the model you trust.

Competitor price trackerEvery day · 08:00
  1. Mon3 price changes · digest sent
  2. TueNo changes
  3. WedNew launch nearby · waiting for approval
  4. ThuApproved · report sent to the head of sales
Budget per run · cap $0.40Spent $0.12
06 · Long-running

Long-running agents

Some work does not end with one reply: watching prices, reviews or tenders, every day for months. Long-running agents work on a schedule, remember what they saw last time, stay within a budget for each run and ask a person when a step needs a decision.

Agentic workflows

Whole processes, built as workflows

Agents, your systems and people in one workflow, with as many steps as the process needs. Build it, test it on real examples, and measure every run once it is live.

spaceplaceAgentsWorkflowsEvaluation LabKnowledgeExecutions (1,284)
Resident requestsVersions (12) · RevertDraftDraft · 11 stepsSandbox · test message✓ Test passed · 4.2 sLive · run 1,284 · completed in 3.9 sRun 1,282 · stopped at «Notify contractor»
YesNo
New messageTrigger · On Message«The kitchen tap is leaking»0.0 s
Request type?Detect IntentRepair request0.2 s
Concierge agentClaude Sonnet 5.5Asked for a photo1.4 s
⇄Create a ticketCall SkillTicket #3180.3 s
⑂Urgent?ConditionNot urgent0.0 s
⇄Alert the managerCall Skill
Which trade?SwitchPlumbing0.0 s
Pick a contractorGPT-6.1AquaFix · Thu 10:000.9 s
Notify contractorHTTP RequestSent0.4 s✕ No answer
Reply in the chatOutputReplied0.2 s
If a step failsError HandlerManager told
Evaluation Lab

Tested on your own cases before it goes live

Your team gathers the questions customers really ask, with the answers you expect. Every new version of an agent runs them all, judge models score each answer, and it goes live only when it clears your bar.

spaceplaceAgentsWorkflowsEvaluation LabKnowledgePass ≥ your bar
Sales assistantEtalon suite 1 · 120 cases · Judges: etalon · faithfulness · relevance · safetyRun all · v4v4 · 120 of 120Run all · v5v5 · 120 of 120
CaseKindVerdict
Is flat 94 still free?SingleAgent answering…Judge scoring…PassPass
Can I pay in instalments?NEW FAILSingleAgent answering…Judge scoring…FailPass
A buyer asks for a viewing, then changes the dayDialogAgent answering…Judge scoring…PassPass
Do you give 10% off for full payment?NEW FAILSingleAgent answering…Judge scoring…FailPass
Which floors have a view of the park?SingleAgent answering…Judge scoring…PassPass
A buyer asks about the state mortgage programmeDialogAgent answering…Judge scoring…PassPass
Compare runs
v3
92%
v4
78%
v5
96%
Your quality bar

2 cases that passed in v3 fail in v4

v5 fixes both · no new failures

Quality gate · waiting for the runBelow your quality bar · not released✓ Passed the quality gatePublishLatest published · v5
01Your own cases

Single questions or whole dialogs, each with the answer you expect or the criteria it must meet.

02A judge for each angle

Is it right against your answer, is it grounded in your sources, is it safe to send.

03A gate before release

A version below your bar does not go live, and every case that used to pass and now fails is named.

Control

People stay in charge, and every step is on record

Guardrails and clear hand-over rules keep the decisions that matter with people, and every run leaves a trace you can open.

Guardrails
  • Prompt Injection Detection
  • Content Filter
  • PII Detection
Platform floor
Escalate when
Low ConfidenceNegative SentimentVIP CustomerKeywords
Take overOlena took overReturn to agentHandoff summary (optional): «Offered 5%, holding No. 94»
08 · Control

People stay in control

Guardrails check what goes into and out of every agent. Clear triggers hand a conversation to a person, who can take over at any moment and give it back with a note.

Request Trace
Execution Waterfall
  1. Context
    0.12 s
  2. Claude Sonnet 5.5
    1.4 s
  3. unit_check_availability
    0.34 s
  4. Guardrail
    0.06 s
  5. Reply
    0.21 s
Token Breakdown · 2,140 in · 180 outModel & Cost · 0.6¢
09 · On record

Every step on record

Each run leaves a trace you can open: what the agent saw, which tool it called, which model answered, how long each step took and what it cost.

Start building

Start building your agent

Three ways in: describe it, build it on the canvas, or bring the agents you already have. Your team builds on its own, or with our engineers.

spaceplaceAgentsWorkflowsEvaluation LabKnowledge
BriefPlanSandboxEvaluatePublish
What should this agent do?

Author agent
Plan
  1. 1Greet the buyer and ask what they are looking for
  2. 2Check which flats are freeunit_check_availability
  3. 3Book a viewing in the CRMcrm_create_lead
  4. 4Pass discount requests to the head of sales
spaceplaceAgentsWorkflowsEvaluation LabKnowledge
NodesDrag to canvas
TriggersTrigger
AIAI PromptAI Chat
LogicConditionSwitchLoop
ActionsCall SkillHTTP RequestWait
New messageTrigger · On Message
Sales agentAI Chat
⑂Asks for a discount?Condition
⇄crm_create_leadCall Skill
Call Skill
spaceplaceAgentsWorkflowsEvaluation LabKnowledge
ManualSpaceTemplatesImport
  1. External Agent · A2A Agenthttps://agents.yourcompany.com/a2a✓ Connected · Pricing agent
  2. MCP ServerYour stock system✓ Connected · 12 tools
  3. Import Skills · SKILL.mdreturns-policy/SKILL.md✓ Imported · 3 skills

Works with your systems. Ready connections to CRMs, ERPs, telephony and messengers, and any other system through its API.

All integrations →

What Spaceplace can do for your business

Send us a process your team runs by hand. In a demo, we show how it would run as a workflow on your own systems.

FAQ

Questions about agents and workflows.

Short answers. Ask us anything else on a call.

Ask us directly →
Do we need developers to build a workflow?Not to start+

Describe the process in plain words and the AI Flow Builder drafts the steps; the Agent Builder drafts an agent from a short brief. Connecting a system that has no ready connection takes a developer, yours or ours.

What happens when an agent isn't sure?It hands over+

You set when an agent passes a conversation to a person: low confidence, a negative tone, a VIP customer or a word you choose. The person can take over at any moment and give it back with a note.

Can a workflow start without a message?On a schedule+

Yes. A workflow can start on a message, on what the customer asks for, or on a schedule, such as every morning at nine. Long-running agents watch prices, reviews or tenders for months, within a budget for each run.

How do we change a workflow once it is live?A new version+

Edit the step, test it in the sandbox and publish it as a new version. Every version is kept, so you can go back to the last one in one click.

Can our own tools and agents take part?MCP and API+

Your tools connect as skills, through an API or MCP. Your systems can call Spaceplace agents through an API, MCP or A2A; we check which protocol versions your agents use during setup.

Do agents remember earlier conversations?Yes+

Each agent keeps a memory of the customer across conversations and channels, so a buyer who wrote in a messenger last week doesn't start again by email.

Documents you need
Privacy

Cookie settings

Choose what this site may use. Necessary cookies keep it working; the rest are off until you switch them on. How we use them

Necessary

Keep the site working and secure, send the forms you fill in, and remember this choice.

Always on
Analytics

Show which pages people read and where they come from, so we can make the site better.

Marketing

Measure which ads bring people here, and show our ads to people who have visited.