LATTIS AGENTS

Run agents that stay inside the lines.

Isolated sandboxes, scoped credentials, streaming tool calls and a full audit trail for every run.

Get API key
Read the docs
Pixel Bloom lattice in warm cream with a single cobalt pixel cluster
Agent run window for refund-review: plan, search_tickets and read_logs finished, issue_refund waiting for approval of a refund over $200
Sandbox sbx_31c file list with /workspace, export_8842.csv, logs and report.md, persists 24 hours
Approval needed at step 4: issue_refund of $200 to workspace 8842, with Approve and Reject

Every run gets its own walls.

An agent run never shares memory, files or credentials with another. Everything it touches is written to the trace.

01

Agent run

support-triage
input · plan · steps
→
02
SBX_31C

Sandbox

filesystem
processes
network policy
secrets
→
03

Tools

your APIs
search
code
→
04

Trace

every step
inputs and outputs
replayable

From script to worker.

State between steps, typed tools, limits you set and a replay for every run.

Persistent sandboxes

Filesystem and processes survive between steps for 24 hours, so long tasks pick up where they stopped.

sbx_31c · running 3 h 12 m
Files panel for sandbox sbx_31c: /workspace, export_8842.csv 12 MB, logs/ with 3 files, report.md updated 2 min ago
Tools panel with 3 registered tools: search_tickets, read_logs and issue_refund, which needs approval

Tool registry

Any HTTP endpoint or Python function becomes a tool, with a generated and validated schema.

lattis tools add ./tools.py

Guardrails

Network policy, spend limits and human approval steps, set per agent and enforced inside the sandbox.

policy: strict · limit: $50
Policy panel for support-triage: network api.internal only, spend limit $50 per run, secrets scoped to sbx_31c, approvals on issue_refund and send_email
Replay panel for trace tr_8f2a with four steps, issue_refund paused, and a Replay from step 3 button

Replays

Every run can be replayed step by step, with the exact inputs and outputs it saw.

trace tr_8f2a · 4 steps
CUSTOMER STORY · HELIO

“Sandboxes gave us the thing we could not build ourselves: isolation we could explain to security.”

Daniel Okafor
Head of Platform, Helio
Read the Helio story
6 weeks
to launch a coding agent
1.2M
sandbox runs a month
99.98%
run completion

Agents, in plain terms.

Read the docs
Q.01

How is a sandbox isolated?

Each run gets its own container with no shared memory, a private filesystem and a network policy you define.

Q.02

Can agents call my internal APIs?

Yes. Register any HTTP endpoint as a tool and scope its credentials to the sandbox.

Q.03

How do approvals work?

Mark a tool as approval required. The run pauses until someone approves it in the console or in Slack.

Q.04

What does a run cost?

Sandbox minutes at $0.012 plus the tokens the agent uses. Approvals and replays are free.

Run your first agent in five minutes.

1M free tokens a month. No card, no sales call, production keys in two minutes.

Get API key
Read the docs
~/my-app
zsh
$ npm install @lattis/sdk
$ export LATTIS_API_KEY=lt_••••••••
$ npx lattis run hello-agent
✓ plan · ✓ search_docs · ✓ answer
● run completed · 3 steps · 0.9 s