How much does Together Code Sandbox cost?

Compute costs $0.0446 per vCPU hour and $0.0149 per GiB of RAM an hour, with access to the current SDK enabled through Together's sales team. Existing CodeSandbox SDK users pay $0.01486 per VM credit, billed by the minute.

Does Together Code Sandbox support MCP and tool calling?

Full under the Agentic Index verification standard. Together Code Sandbox publishes public evidence of Model Context Protocol (MCP) support, tool calling, and API or SDK extensibility for connecting agents to external tools. Evidence basis: Together's CLI and Python and TypeScript SDKs for sandboxes, generated from its OpenAPI specifications, cover snapshot builds, sandbox lifecycle, command execution with streaming, files, file watching, ports, tags and retries. Together offe. Graded from public evidence only against the Agentic Index 14 point capability taxonomy. No vendor pays for placement.

Does Together Code Sandbox have SOC 2, SSO and role based access control?

Full under the Agentic Index verification standard. Together Code Sandbox publishes public evidence of enterprise security and identity controls, covering a compliance posture such as SOC 2 or ISO 27001, single sign on (SSO or SAML), and role based access control (RBAC). Evidence basis: Together AI has completed a SOC 2 Type 2 examination, supports HIPAA and signs business associate agreements, without naming the products the report covers. Organizations have admin and developer roles, projects have admin and editor roles. Graded from public evidence only against the Agentic Index 14 point capability taxonomy. No vendor pays for placement.

Can you bring your own model or route between providers with Together Code Sandbox?

Full under the Agentic Index verification standard. Together Code Sandbox publishes public evidence of bring your own model and bring your own key (BYOK), and of routing tasks between models or providers rather than being locked to one. Evidence basis: The customer chooses the models its agents use. Code in a sandbox can call any model, and the same Together account serves models through Together's serverless and dedicated inference, which Together pairs with the sandboxes. Graded from public evidence only against the Agentic Index 14 point capability taxonomy. No vendor pays for placement.

Back to vendors

Together Code Sandbox

Q: Who is Together Code Sandbox for?

AI product teams, agent builders, AI research teams

Also known as: Together Sandbox, CodeSandbox SDK, together-sandbox, @codesandbox/sdk

Visit site

Agent infrastructurefirst party productVerified 2026-11-05

Entry priceUsage based at $0.0446 per vCPU hour and $0.0149 per GiB of RAM an hour, with access to the current SDK requested through Together's sales team.Full pricing detail

Together Code Sandbox is Together AI's service for isolated sandboxes built from Docker image snapshots, for agent tool calls, model generated code and reinforcement learning rollouts, and the successor to the CodeSandbox SDK.

Together Code Sandbox is Together AI's service for running isolated sandboxes for agents, model generated code and reinforcement learning rollouts, built by the CodeSandbox team Together AI acquired in December 2024. A team defines an environment as a snapshot built from a Docker image or Dockerfile on Together's servers, then starts sandboxes from it to run commands, read and write files and serve ports. Together AI also sells model inference, fine tuning and GPU clusters, and the sandboxes sit on the same account as its models.

The current Together Sandbox SDK comes as a CLI and Python and TypeScript SDKs that sign in with a Together API key. Sandboxes take from 0.1 to 16 vCPUs with 1 to 8 GB of memory per vCPU, stream command output and file changes back to the caller, and run until terminated or until a TTL ends them, with the filesystem saved as a new snapshot on termination when asked. Organizations get the new SDK enabled through Together's sales team. The earlier CodeSandbox SDK, which hibernates running VMs with their memory, forks them and resumes them from snapshots, is now deprecated for existing users, though Together still promotes those capabilities for Sandbox.

Compute costs $0.0446 per vCPU hour and $0.0149 per GiB of RAM an hour, and existing CodeSandbox SDK users buy VM credits at $0.01486 each. Together AI has completed a SOC 2 Type 2 examination and supports HIPAA with business associate agreements, organizations have admin and developer roles with admin and editor roles on each project, and SAML or OIDC single sign on comes with Scale and Enterprise accounts. The current SDK offers no region choice, while the CodeSandbox SDK runs in data centers in the US, Asia Pacific and Europe.

Together Code Sandbox fits teams already running models on Together AI who want sandboxes for agent tool calls, generated code or reinforcement learning rollouts from the same vendor and bill. It fits less well for buyers who want self serve sign up for the current SDK, sandboxes that wake on incoming traffic, a browser for agents, or a settled product line, since the hibernation and forking Together promotes belong to an SDK it now treats as deprecated.

Vendor details

Canonical URL

https://www.together.ai/code-sandbox

Funding status

Together Code Sandbox is a product of Together AI, which acquired CodeSandbox in December 2024.

Company status

first party product

Use cases & customers

Primary use cases

code execution for LLM outputagent sandboxesreinforcement learning rolloutscoding agents

Target customers

AI product teamsagent buildersAI research teams

Deployment options

SaaSAPI

Integrations

The Together Sandbox CLI and Python and TypeScript SDKs, generated from Together's OpenAPI specifications, manage snapshots and sandboxes with a Together API key, and the legacy CodeSandbox SDK package serves existing users. Together's guides cover running Harbor benchmark jobs and a Tinker with Harbor recipe in its sandboxes, and the same account reaches Together's model inference, fine tuning and GPU clusters.

In practice

Your model writes code that has to run somewhere it can't touch your own systems. Build a snapshot from your Dockerfile on Together, start a sandbox from it for each request, run the code, read the output and let a TTL clean the sandbox up.

You're running reinforcement learning rollouts that each need the same starting environment. Build one snapshot with your tools installed and start as many sandboxes from it as the run needs, each sized from a tenth of a vCPU up to 16.

Your agent's session needs to carry its files into the next run. Terminate the sandbox with a snapshot, give that snapshot an alias, and start the next sandbox from the alias with the files where the agent left them.

Sources & related URLs

Related / legacy domains

https://codesandbox.io/sdk https://docs.together.ai/docs/together-code-sandbox

Research sources

https://www.together.ai/code-sandbox https://www.together.ai/pricing https://docs.together.ai/docs/together-code-sandbox https://docs.together.ai/docs/together-code-sandbox-legacy.md https://github.com/togethercomputer/together-sandbox https://codesandbox.io/sdk https://together.ai/blog/soc-2-compliance https://docs.together.ai/docs/roles-permissions.md https://docs.together.ai/docs/sso.md https://together.ai/blog/codesandbox-acquisition-together-code-interpreter

Agentic Index coverage score

4.5 / 14 capabilities · 32%

Integrations & Tool Calling	Not documented
No connectors, tool catalog, secrets store or credential brokering for outside systems comes with the sandboxes. Commands run with environment variables the caller passes on each execution, and an agent's code inside the sandbox reaches other systems on its own. Sourcegithub.com/togethercomputer/together-sandbox/blob/main/docs/python-sdk.mdread 2026-11-05
Workflow Orchestration	Not documented
There is no workflow logic in the product. The SDKs start sandboxes, run commands, retry failed API calls with backoff and list sandboxes by status and tag, but sequencing, branching and routing an agent's steps stay in the customer's own code. Sourcegithub.com/togethercomputer/together-sandbox/blob/main/docs/python-sdk.mdread 2026-11-05
Knowledge Grounding & RAG	Not documented
No retrieval index or knowledge base comes with the sandboxes. Files an agent needs are built into the snapshot or written in at run time, and any retrieval runs in the customer's own code. Sourcedocs.together.ai/docs/together-code-sandboxread 2026-11-05
Human Oversight & Guardrails	Not documented
Work in a sandbox has no approval step, consent checkpoint or escalation to a person. Commands can run in a terminal that accepts input, but nothing holds an agent's action until someone reviews it. Sourcegithub.com/togethercomputer/together-sandbox/blob/main/docs/python-sdk.mdread 2026-11-05
Security, Identity & Governance	Full
Together AI has completed a SOC 2 Type 2 examination, supports HIPAA and signs business associate agreements, without naming the products the report covers. Organizations have admin and developer roles, projects have admin and editor roles with open, closed or private visibility, external collaborators can join single projects, and SAML or OIDC single sign on with Okta, Microsoft Entra or Google Workspace comes with Scale and Enterprise accounts. Sandboxes run under the account's Together API key, with no sandbox specific permissions among the roles. Sourcedocs.together.ai/docs/roles-permissions.mdread 2026-11-05
Observability & Auditability	Not documented
Sandboxes report their status, and command output, file changes and open ports stream back to the caller, with output resumable from a sequence number. No logs, metrics, event history or audit trail of sandbox activity comes with the product. Sourcegithub.com/togethercomputer/together-sandbox/blob/main/docs/python-sdk.mdread 2026-11-05
Memory & State Persistence	Partial
Work carries forward through snapshots. A sandbox can be terminated with a snapshot of its filesystem that takes an alias and starts the next sandbox, and many sandboxes can start from one snapshot. Together also promotes hibernation that keeps disk and memory and forks of hibernated sandboxes, which belong to the CodeSandbox SDK it now treats as deprecated. This is machine state the developer manages, with no memory layer of stated scope. Sourcedocs.together.ai/docs/together-code-sandboxread 2026-11-05
Deployment & Data Residency	Partial
The CodeSandbox SDK runs in data centers in the US, Asia Pacific and Europe, but the current SDK has no region parameter or other way to choose where a sandbox runs. Together's private networking and VPC deployments, including EU regions, cover its dedicated inference endpoints rather than sandboxes. Sourcecodesandbox.io/sdkread 2026-11-05
Prebuilt Agents / Templates / Packs	Partial
Starting points are snapshots and guides. Snapshots build from any public Docker image or Dockerfile and take reusable aliases, the legacy SDK ships a Universal template, and Together's guides cover running Harbor benchmark jobs and a Tinker with Harbor recipe in sandboxes. These are environments a developer assembles, not prebuilt agents a business adopts. Sourcegithub.com/togethercomputer/together-sandboxread 2026-11-05
Triggers & Channel Coverage	Not documented
Sandboxes start only when the customer's code or the CLI creates them. There is no schedule, webhook or wake on inbound traffic, and a TTL only ends a sandbox. Sourcedocs.together.ai/docs/together-code-sandboxread 2026-11-05
Model Flexibility & Routing	Full
The customer chooses the models its agents use. Code in a sandbox can call any model, and the same Together account serves models through Together's serverless and dedicated inference, which Together pairs with the sandboxes. Sourcetogether.ai/code-sandboxread 2026-11-05
APIs / SDKs / MCP Extensibility	Full
Together's CLI and Python and TypeScript SDKs for sandboxes, generated from its OpenAPI specifications, cover snapshot builds, sandbox lifecycle, command execution with streaming, files, file watching, ports, tags and retries. Together offers no MCP server for sandboxes. Sourcegithub.com/togethercomputer/together-sandboxread 2026-11-05
Testing, Debugging & Optimization	Not documented
There are no test cases, scoring or quality gates for an agent's behavior. Sandboxes host evaluation and reinforcement learning runs, with Together's Harbor guides for benchmark jobs, and the datasets and scoring belong to those harnesses. Sourcegithub.com/togethercomputer/together-sandboxread 2026-11-05
Browser / Computer-use	Not documented
No browser or desktop for an agent to drive comes with the sandboxes. Environments are built from Docker images, and an agent that needs a browser installs and drives one itself. Sourcedocs.together.ai/docs/together-code-sandboxread 2026-11-05

The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded

Pricing

Usage based at $0.0446 per vCPU hour and $0.0149 per GiB of RAM an hour, with access to the current SDK requested through Together's sales team.

vCPU hours and GiB of RAM hours per sandbox, or VM credits per hour by VM size on the legacy SDK

Included quota

The current SDK comes with no included usage. On the legacy CodeSandbox SDK the free Build plan allows 10 concurrent VMs, Scale allows 250 and Enterprise is custom.

What is public

Together publishes per vCPU hour and per GiB of RAM hour rates for Code Sandbox, and the legacy CodeSandbox SDK's credit price, VM sizes and concurrency limits.

Billing mechanics

Each sandbox bills by its vCPUs and memory for the hours it runs, from creation until it is terminated or its TTL ends. Legacy CodeSandbox SDK runtime bills in VM credits by size, counted in whole minutes.

Cost watchouts

Sandboxes on the current SDK run until terminated unless a TTL or the CLI's --rm flag ends them, so idle sandboxes keep billing. Snapshot storage has no public price.

Variable cost rationale

Cost follows the vCPU and memory of every sandbox for every hour it runs, and sandboxes keep running until terminated unless a TTL is set, so large rollouts or forgotten sandboxes scale the bill directly.

Additional watchouts

A default sandbox with 1 vCPU and 2 GiB of RAM costs about $0.07 an hour, and a 4 vCPU, 8 GiB sandbox about $0.30 an hour.

Overage / add-ons

Usage bills at the per hour rates, and the legacy SDK bills runtime in whole minutes, rounded up.

Sales call required

Mixed (some tiers require a call)

Free / trial

No free tier for the current SDK, while the legacy CodeSandbox SDK has a free Build plan limited to 10 concurrent VMs

Lowest paid plan

Usage at $0.0446 per vCPU hour and $0.0149 per GiB of RAM an hour

Commercial notes

Together AI sells Code Sandbox alongside its inference, fine tuning and GPU clusters on one account, and new organizations request access to the current SDK through sales.

Key ambiguities

Access to the current SDK goes through Together's sales team, and the rates come without plan tiers, minimums or committed use terms.

Missing data

No snapshot storage price, no plan tiers or minimums for the current SDK, and no end date for the legacy SDK's pricing.

Agentic Index verified 2026-11-05

Official pricing page Docs

Data confidence: medium

Head to head

Together Code Sandbox vs boxdTogether Code Sandbox vs E2B