Tembo

Platform · Last verified 2026-09-30

TL;DR

Tembo is a hosted platform that runs existing coding agents (Claude Code, Codex, Cursor, Amp, OpenCode, Pi) in isolated cloud sandboxes, one VM per session. Sessions start from issues, alerts, chat mentions, schedules or the API and return pull requests. Tembo the company maintains it; a self-hosted option exists. It suits teams that want background agent runs across repositories.

Key facts

Tembo key facts. Data as of 2026-09-30.
Type	Platform
Languages / SDKs	TypeScript
License	Proprietary
Pricing model	Hosted, paid
Orchestration pattern	Other
Repository	No public repository (closed source)
Website	www.tembo.io
Documentation	docs.tembo.io
Last verified	2026-09-30

Key features

Harness- and model-agnostic: each agent or session picks Claude Code, Codex, Cursor, OpenCode, Amp, Pi or fx, with a default per agent and an override per session. (source)
Every session runs in its own ephemeral Linux VM sandbox (nano to XL sizes) that is destroyed when the session ends; cached project environments are the exception. (source)
Agents can be started on a schedule (hourly to monthly) or by events such as a new Sentry error, a Linear issue, a GitHub pull request or a Slack mention. (source)
Quickstart flow: connect GitHub, GitLab or Bitbucket, create a session with a chosen harness, review the diff in the Changes tab, then open a pull request and mention @tembo in comments to request changes. (source)
Built-in and custom MCP servers (stdio, HTTP, SSE, OAuth) give agents tools during a session, and a hosted Tembo MCP server exposes the public Tembo API to external MCP clients. (source)
A public API and a TypeScript SDK create sessions, agents, schedules, triggers and runs programmatically. (source)
Bring your own model access: provider API keys, Amazon Bedrock and GCP Vertex AI credentials (paid plans), or a ChatGPT or Claude subscription, alongside Tembo-managed inference. (source)
A self-hosted edition runs the platform as one NixOS-based instance in infrastructure you control. (source)

Architecture and orchestration pattern

Pattern: Other

Tembo is a runner and control plane for other vendors' coding agents, not a model or agent loop of its own. A session is a dedicated VM sandbox with the repository cloned, a terminal, files and optional connected tools; the chosen harness (for example Claude Code with a Sonnet model, or Codex) runs inside it. Tembo handles orchestration, session context, sandbox execution and pull request automation, while model access can be Tembo-managed or the customer's own key or subscription.

Multi-agent use is by composition of independent sessions. Different tasks can use different harnesses, work can be triggered in parallel by schedules and events, and a single task can open coordinated pull requests across several repositories. The pricing page says the Max plan is for teams with more concurrent tasks. The vendor's llms.txt also mentions a MAX mode for side-by-side comparison of agent outputs, but no docs page on it was readable. No coordinator agent that splits work among sessions is described.

State lives in the session record (logs, diffs, artifacts) and in reusable Projects, which cache repositories and dependencies so new sessions start faster. Rule files and agent skills in the repository carry conventions between sessions.

Human in the loop

The main control point is review of the result: the session's Changes tab shows the diff, and the person clicks Open PR or asks Tembo to open one, then requests revisions by mentioning @tembo in pull request comments. The vendor site says users can approve, reject or redirect from Slack, Linear or GitHub, and live session logs stream to the dashboard and those tools. MCP write operations should be reviewed by the person before approval in their client, according to the API MCP page. Whether a running session can be paused for approval of individual tool calls is unknown from the pages read.

Harnesses it can drive

Claude Code (evidence)
Codex (evidence)
OpenCode (evidence)

Protocols

MCP, A2A and AG-UI support for Tembo. See the full matrix.
Protocol	Support	Note
MCP	Yes evidence checked 2026-09-30	Client: agents use built-in and custom MCP servers (stdio, http, sse, OAuth) during sessions. Server: a hosted Tembo MCP server at mcp.tembo.io exposes the Tembo API to external clients (docs.tembo.io/api/mcp).
A2A	Unknown checked 2026-09-30	No mention of A2A or Agent2Agent in www.tembo.io/llms.txt or llms-full.txt; the docs index (docs.tembo.io/llms.txt) lists no A2A page. github.com/tembo/tembo contains only a stub README.
AG-UI	Unknown checked 2026-09-30	No mention of AG-UI in www.tembo.io/llms-full.txt or the docs index, and Tembo is not listed in the AG-UI protocol README.

Best for

Teams that want tickets, Sentry errors or Slack requests turned into pull requests by a background agent in a full VM. (shortlist)
Running the same task or different tasks on different agent harnesses (Claude Code, Codex, Cursor, OpenCode) from one dashboard, with per-task choice. (shortlist)
Scheduled maintenance work such as dependency upgrades, docs sync and security scans across many repositories.
Organisations that need to run the platform inside their own infrastructure (self-hosted edition). (shortlist)

Not for

Users who want a local, offline agent on their own machine; sessions run in Tembo-managed or self-hosted cloud VMs.
Anyone who needs an open-source platform: the product is proprietary and github.com/tembo/tembo holds no source.
Workflows that need agents to negotiate or hand off to each other; the docs describe independent sessions.

Quickstart

Sign in at https://app.tembo.io, connect GitHub, GitLab or Bitbucket under Integrations, then click New Session.

Install not yet verified by this site. What this means

# From the official quickstart (web dashboard)
1. https://app.tembo.io/integrations -> connect GitHub, GitLab or Bitbucket; select repositories
2. https://app.tembo.io -> New Session -> pick repositories and an agent harness
3. Prompt with what, where, why and how you will know it is done:

   Fix memory leak in workers/job-processor.ts
   Heap grows from 512MB to 4GB after ~6 hours; likely cleanup logic near line 156.
   Done when: memory is stable over 24+ hours and no queued jobs are lost.

4. Review the diff in the Changes tab, then Open PR
5. Comment @tembo on the pull request to request changes
6. https://app.tembo.io/agents -> pick a template to run on a schedule or trigger

Common pitfalls

The two official pricing pages differ on the Free plan: the marketing page shows a 14-day free trial for one user, the docs page says a $10 one-time allowance and up to 3 users. Check the current plan in Settings, Billing.
Free sandboxes are capped at the micro size; Pro allows up to medium and Max up to XL.
A tag is required in PR comments, GitHub issues and Slack (@tembo); dashboard sessions and Linear or Jira issues assigned to Tembo need no tag.
Cloud BYOK (Amazon Bedrock, GCP Vertex AI) needs a paid plan.
Included monthly allowance expires each billing cycle and does not roll over; on Free, new sessions are blocked when the allowance is used up.
Self-hosted Kubernetes and GCP support is early access; self-hosted is a single NixOS-based instance.

Official quickstart

Pros

Works with six or more coding-agent harnesses and lets you set the harness per agent or per session. (source)
Each session gets a dedicated VM, so no two sessions share a machine, with nested virtualization for Docker workloads. (source)
BYOK and subscription options mean model inference can cost $0 through Tembo, leaving only VM compute against the allowance. (source)
Triggers cover schedules and events from GitHub, GitLab, Sentry, Linear and Slack, so runs need not be started by hand. (source)
Offers a self-hosted edition and an API with a hosted MCP server for external clients. (source)

Cons

Proprietary and metered: usage draws on a dollar allowance for VM compute and Tembo Gateway model use, and Free stops new sessions when it runs out. (source)
The Free plan is described differently on the pricing page (14-day trial, one user) and in the docs (one-time allowance, up to three users). (source)
Self-hosted is a single-instance deployment, and Kubernetes and GCP support are early access. (source)
Sandboxes are ephemeral: no repository contents or working state persist after a session, except cached project environments. (source)

Alternatives

FAQ

Does Tembo support MCP?

Yes. Agents can use built-in and custom MCP servers during sessions over stdio, HTTP or SSE, with OAuth support, and Tembo hosts an MCP server at mcp.tembo.io that exposes its API to other MCP clients.

Which coding agents does Tembo run?

Claude Code, Codex, Cursor, OpenCode, Amp, Pi and fx, chosen per agent or per session. Tembo runs them in its sandboxes rather than replacing them.

How much does Tembo cost?

The pricing page lists Pro at $60 per month and Max at $200 per month, each with an equal dollar usage allowance, plus a custom Enterprise plan; the Free plan terms differ between the marketing page and the docs page (read 2026-09-30). VM compute is billed at $0.0403 per vCPU-hour plus $0.0130 per GiB-hour of RAM per the docs.

Is Tembo open source?

No. The product is proprietary. The github.com/tembo/tembo repository has only a placeholder README, so no source is listed here.

Sources

Unknown fields: A2A and AG-UI: searched www.tembo.io/llms.txt, llms-full.txt and the docs index for A2A, Agent2Agent and AG-UI and found nothing; the github.com/tembo/tembo repository has only a stub README so no code search was possible. HITL: the pages read do not say whether an individual tool call inside a running session can be paused for approval. The MAX mode (side-by-side comparison) is mentioned in llms.txt but docs.tembo.io/features/max did not render as readable text, so it is not used as a feature. Languages: only the TypeScript SDK (@tembo-io/sdk) is named; its install command was not verified from a docs page, so the quickstart uses the web dashboard. Free plan terms conflict between www.tembo.io/pricing (14-day trial, one user) and docs.tembo.io/resources/pricing ($10 one-time allowance, up to three users); the record does not state Free terms as fact.

Something wrong or out of date, or do you maintain Tembo and want this page removed? Report a correction or request removal.