For IT and AI Teams

Ground Your Agents In The Expertise Of Your Operations Teams

Agent builders need more than integrations and prompts. They need expert verification from the teams doing the work. Interloom is the bridge: you connect agents to systems, keep every key and action under control, and turn reviewed operational decisions into a trusted memory layer for the next case.

Get Started Read The Docs Permissions · Security · Auditability · Cost control · BYOK

Connect Your Agents In Under A Minute

Run the Interloom CLI with npx -y @interloom/cli, save an API key with auth login, and your agents can use the same credentials for CLI calls or the local MCP server. The MCP server runs directly as npx -y @interloom/cli mcp and exposes the CLI's API operations as tools.

# Save an API key for app.interloom.com
npx -y @interloom/cli auth login

# Or pipe the key for CI and agents
echo "$INTERLOOM_API_KEY" | npx -y @interloom/cli auth login

# Verify the active user and organization
npx -y @interloom/cli auth status

# Create your first case
npx -y @interloom/cli cases create \
  --title "Northwest claim, water damage" \
  --description "Customer reports water damage after heavy rain"

curl https://app.interloom.com/api/v1/public/cases \
  -H "Authorization: Bearer $INTERLOOM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "space_id": "0196d3b9-0c9d-7cf2-9c36-2a8f0bb0f3aa",
    "title": "Northwest claim, water damage",
    "description": "Customer reports water damage after heavy rain. Review photos, policy coverage, and next steps.",
    "external_id": "claim-NW-1042",
    "due_at": "2026-07-10T09:00:00.000Z",
    "tags": ["claims", "water-damage", "urgent"]
  }'

# 201 Created
{
  "id": "3fa85f64-5717-4562-b3fc-2c963f66afa6",
  "type": "CASE",
  "created_at": "2026-07-03T12:32:17.908Z",
  "updated_at": "2026-07-03T12:32:17.908Z",
  "parent": {
    "id": "0196d3b9-0c9d-7cf2-9c36-2a8f0bb0f3aa",
    "type": "SPACE",
    "url": "/api/v1/public/spaces/0196d3b9-0c9d-7cf2-9c36-2a8f0bb0f3aa"
  },
  "title": "Northwest claim, water damage",
  "status": "open",
  "description": "Customer reports water damage after heavy rain. Review photos, policy coverage, and next steps.",
  "due_at": "2026-07-10T09:00:00.000Z",
  "external_id": "claim-NW-1042",
  "tags": ["claims", "water-damage", "urgent"],
  "thread": {
    "id": "0197f1bc-7369-7343-ae04-31dc04cb204d",
    "type": "THREAD",
    "url": "/api/v1/public/threads/0197f1bc-7369-7343-ae04-31dc04cb204d"
  }
}

# Uses your saved CLI credentials and stdio by default
npx -y @interloom/cli mcp

# Generic MCP client config
{
  "mcpServers": {
    "interloom": {
      "command": "npx",
      "args": ["-y", "@interloom/cli", "mcp"]
    }
  }
}

Expert-Verified Memory

Build The Bridge Between Agents And Experts

Your agents improve only when the right people verify what should be remembered. Interloom lets operations experts review proposed changes to decisions, procedures, and context before they become part of the memory layer. Think pull requests for the context graph: reviewed, attributed, reversible.

Expert verification

Ops experts approve the decisions, edge cases, and corrections your agents should learn from before they enter memory.

Pull requests for memory

Every proposed update to the context graph is reviewed like a diff: what changed, why it changed, and who signed off.

Grounded agents

Builders ship agents that reason from reviewed precedent, not generic prompts, stale documentation, or unverified traces.

Memory pull request

Claims decision · water ingress precedent

Expert review

Case WI-8842 · Northwest claim · proposed by triage-bot

Update claims memory after expert resolution

The agent found a similar appliance leak claim, but Maria O. corrected the outcome and linked it to a prior successful resolution. This change will update the context graph used for future water damage decisions.

Decision: Escalate to senior adjuster because tenant liability is unclear.

Decision: Approve partial cover under water ingress precedent WI-2217.

Reason: Missing precedent for appliance leak in mixed-use unit.

Reason: Matched prior case WI-2217, reviewed by Maria O., same deductible pattern.

Memory: Link appliance leak + partial cover + deductible dispute to Claims Water Damage procedure.

Full Visibility

See Exactly What Your Ops Teams Are Doing

The same surface the work runs on is the one you watch it from. Throughput, lead time, cost per turn, success rate, broken down by space, by agent, by model, and by case. No warehouse to feed, no separate analytics product to wire up.

Space · Leasing Last 30 days

Throughput 25.924/mo

Lead time 12,3d

Handle time 6,6d

Success rate 98,8%

Turns 284.990

Total cost €107,432

Avg cost / turn €0.3770

Avg duration 24s

By model

openai/gpt-5.5 turns118.420 cost€42,631 avg dur28.4s

openai/gpt-5.4-nano turns92.160 cost€8,294 avg dur11.2s

google/gemini-2.5-flash turns44.870 cost€47,054 avg dur36.8s

openai/gpt-5.4 turns29.540 cost€9,453 avg dur22.6s

Top consuming cases

Handover, Thornfield Pharma, Northfields Unit 9 turns86 cost€37.92

New Lease: Orbital Freight, Building 3 Unit 2 turns74 cost€31.06

30-Day Check-In: Helix Metals, Brightfield Unit 4 turns68 cost€27.41

Move-Out, Quanto Robotics, Building 2 turns61 cost€24.18

The Control Plane

You Decide Who Can Do What

More agents make an organization harder to govern, not simpler. Each one carries its own keys, permissions, and access, and that complexity compounds with every person who builds one. The hard part isn’t storing the keys; it’s keeping control as many people manage many agents.

So we route control through a structure that already works: your org chart. Agents join as first-class peers in an agentic org chart and inherit the reporting lines, scopes, and approvals you already run. Keys and permissions follow the same chain of authority as the people beside them.

Permissions

Decide which agent can enter which space, which tool a given agent may call, and which key each tool is allowed to use. Least privilege by default.

Security

Bring your own keys. Scope every credential to a single tool, rotate without a redeploy, and keep secrets out of prompts and scripts.

Governance

Models, clouds, and agent configs stay owned by IT and AI teams. Ops teams operate inside the guardrails you set, not around them.

Auditability

Every tool call, stage transition, and agent decision is recorded as a typed activity. Replay any case end to end.

Cost control

Watch cost per turn, per agent, per model, per case. Cap spend and route expensive stages to cheaper models when quality allows.

Bring Your Own Keys

Keys are about to be one of the hardest problems in agentic operations. In Interloom each secret is scoped to a single tool, bound to the agents allowed to use it, and rotated without touching a script. You decide which tool can use which key, and which agent can use which tool.

Secrets · scoped per tool

OPENAI_API_KEY GPT inference for the agent harness

GEMINI_API_KEY Gemini web-search grounding and reranking

SLACK_BOT_TOKEN Inbound Slack integration

STRIPE_SECRET_KEY Customer billing automation

KV_REST_API_TOKEN Upstash KV, live thread storage

Staging And Production

A Controllable Environment, Like Real Software

Build and trial an agent in a sandbox, run it in shadow mode against live cases to compare its decisions to the manual ones, and promote it to production only when the numbers hold. The same procedures, agents, and audit trail move with it.

01 · Trial Build & shadow

02 · Staging Verify on real cases

03 · Production Promote when proven

Interoperable With Forge

Build Any Integration In Minutes

Forge is where your teams build bespoke tools, scripts, skills, agents, and surfaces for the organization. Wrap any API as a tool, promote a script to a tool, and bind it to the agents and spaces that should use it, in minutes, not sprints.

Tools

Wrap any API as a typed tool an agent can call.

Scripts

Deterministic logic for the steps that must not drift.

Skills

Reusable capabilities composed from tools and prompts.

Agents

Configured field workers with their own scope and keys.

Surfaces

Custom UI a tool renders into the workspace.

Integrations

First-party connectors to the systems you already run.

Grounded Decisions

A context graph behind every decision. Agents reason over a living graph of cases, precedent, and relationships, not generic training data. Explore the context graph →

One API Surface

Every action is a typed, replayable trace. Introspect the schema, create cases, and stream activity over REST or MCP, with webhooks for everything in between.

GET /v1/threads/th_44/activities

{ "type": "stage_transition", "from": "intake", "to": "review", "actor": "act_triage_ai" }
{ "type": "tool_call", "name": "documents.extract", "status": "ok", "ms": 812 }
{ "type": "comment", "actor": "Maria O.", "visibility": "public" }
{ "type": "status_change", "field": "status", "to": "blocked" }
{ "type": "stage_transition", "from": "review", "to": "decision", "actor": "Maria O." }

RESTful Cases

POST /v1/cases

Create, transition, and assign cases with a predictable REST surface.

Activity Stream

GET /v1/threads/:id/activities

Subscribe to a typed stream of comments, transitions, and tool calls.

Trace Data Model

tool_call · stage_transition

Every agent step is a typed, replayable trace record.

Webhooks

case.resolved · stage.completed

Push events to your services the moment state changes.

MCP Server

npx -y @interloom/cli mcp

Drop Interloom into any agent runtime over stdio, backed by your saved CLI config.

Stop rebuilding state machines, audit logs, and key vaults for every agent. Give your ops teams a governed environment you control end to end.

Book A Demo Read The Docs