Cloud workspaces. Explicit access.
Code and file tools run in isolated remote workspaces scoped to a run. Connect the repositories and services the agent needs, without handing it an employee’s machine.
Explore execution boundaries →Open architecture. Enterprise control.
The AI agent harness built for enterprise.
Run capable agents on infrastructure you control. Choose your models. Keep code execution in isolated cloud sandboxes and private context tied to its source. Mesh gives your organization an agent harness it can inspect, self-host, and make its own.
Connected todaySlackTelegram
YOUR ORGANIZATIONARCHITECTURE / ILLUSTRATION
MeshThe agent harness you operate
Agent identitiesSource permissionsRun records
INFERENCEYour choice of models
Provider keys or a private
compatible endpoint
EXECUTIONIsolated workspaces
Remote code & files
Scoped to the run
Self-hosted or Texture-hostedMIT-licensed core
YOUR INFRASTRUCTUREYOUR MODELS. YOUR DATA. YOUR CONTROL.EXPLORE ↓
The enterprise starting point
Give agents a job.
Define their boundaries.
Putting an agent to work should be an infrastructure decision: where it runs, which systems it can reach, and whose information it can use. Those choices belong to your organization.
Keep your options open.
Bring your own provider keys. Choose supported hosted models or connect a self-hosted, OpenAI-compatible endpoint for open-weight inference on your own hardware.
Choose your inference →Context carries permissions.
Authorship and source audiences follow conversations and memory. Private context is checked against the destination audience before it enters a turn and again before delivery.
See how private context stays private →Your deployment. Your choice.
An MIT-licensed core you can inspect and self-host, independent of Texture’s hosting service. Evaluating an airgapped environment? Start with the deployment requirements.
Self-hosting & airgapped deployments →Mesh is in beta. The MIT-licensed source is currently access-restricted while we prepare its public release. Source access & deployment status →
Built for the way teams work
You don’t just prompt it.
You onboard it.
Give it a job in your company. Shape how it communicates. Bring it into the channels where the team works, with the tools and context it needs to contribute.
Call and response is an exchange. A virtual employee has a place in the organization. Your team works with the same recognizable agent, building useful memory and shared history across conversations.
Why not just ChatGPT, Claude, or Gemini?ONE AGENT, OVER TIMEILLUSTRATION
NovaYour team’s release partner
SAME IDENTITY- SHAPE ITS CHARACTER“Be direct. Lead with risks. Show your evidence.”
A persona you write and can revise.
- BUILD SOME HISTORY“Remember that Ada owns launch communications.”
A saved memory, attributed to its source.
- PICK UP THE THREAD“Who should we loop in for the release?”
“Loop in Ada—she owns launch communications.”
01 / The social part of softwareTry a conversation
Good coworkers know
when to join in.
A shared channel contains questions, side conversations, and messages an agent should leave alone. Mesh treats those as different situations. Pick one and see why.
# release-roomIllustrative scenario
MESH DECISIONEngage
Nova was directly mentioned. Ada is the requester; the room is shared context.
Decision signal
direct mention02 / A little personal space
Work together.
Think separately.
Your research agent and coding agent can live in the same Mesh. Each owns its actors, history, and memory. Private sources keep their audience restrictions across recall, amendments, and summaries.
See how perspectives workONE ROOMTWO PERSPECTIVES
AAda# project-roomSSam
Nova
ActorsHistoryMemory
Knows Ada’s release context
Orbit
ActorsHistoryMemory
Knows Sam’s research context
Built to be understood
“Why did it do that?”
should have an answer.
Mesh records participation decisions, prompt snapshots, model calls, tool observations, and run outcomes. The explanation starts with the records behind the response.
Open the anatomy of a turn03 / Choose where it runsThen choose AI and execution
Own the deployment.
Choose who operates it.
Self-host the software, or run a dedicated hosted instance. For hosted Mesh, AI and execution are separate choices: bring your own keys, or use ours with pass-through billing where available.
YOUR INFRASTRUCTURE · SOURCE ACCESS ON REQUEST
Explore the project ↗Self-host Mesh
Control the infrastructure, model endpoints, data, and configuration.
FOR TECHNICAL TEAMS · BETA
$99 / instance / month ↗Hosted Mesh
We host your Mesh. You configure the agents and choose AI and execution.
Inference
Managed AI (OpenRouter)Available
Choose it when you launch an instance. Texture provisions an OpenRouter key for that instance and installs it in Mesh, within a monthly spending limit you set in the console.
OpenRouter usage at the provider's rates plus a 15% management fee on that usage. Taxes are separate. Paid from prepaid usage funds, bought before first use.
Your own model keyAvailable
Add an OpenRouter or Ramp Router API key inside your Mesh instance. A ChatGPT or Claude subscription is not an API key.
Billed directly by your provider. No Texture markup.
Execution
Managed execution (Modal)Available
Choose it when you launch an instance. Code and repository tools run in Modal sandboxes on Texture’s Modal account, through a revocable, rotating credential issued to that instance only.
Modal sandbox time at Modal’s list price, metered per second, plus a 15% management fee on that cost. Taxes are separate. Per sandbox hour: Small $0.1770, Standard $0.3816 (Modal’s list price plus the fee). Paid from prepaid usage funds, bought before first use.
| Sandbox size | Modal’s price | You pay |
|---|---|---|
| Small: 1 CPU core, 512 MiB memory | $0.1539/hour | $0.1770/hour |
| Standard: 2 CPU cores, 2 GiB memory | $0.3318/hour | $0.3816/hour |
Your own Modal or Fly accountAvailable
Optional. Add a Modal token or a Fly token in your Mesh instance’s settings for code and repository tools.
Billed directly by Modal or Fly. No Texture markup.
Configure connectors and create agents inside your Mesh instance. For Slack, connect management credentials once for the instance, then give each agent its own Slack app.
The 14-day trial covers the hosted instance only and starts at Checkout for your first eligible instance. Additional instances are billed without a trial. AI inference and execution are not part of the trial. Managed services are paid from prepaid usage funds, bought before first use; your own provider keys are billed by those providers.