Agents

Run coding agents in the background, on a schedule or triggered by events. A Sentry error gets fixed within seconds. Docs stay in sync with your code. Security vulnerabilities get caught before they become problems. Each agent runs in a secure sandbox with full codebase and repository access and your configured MCP servers.

Agent harness

Tembo is harness and model agnostic. For each agent you configure, you pick the harness (and model) that runs in the session it spawns, for example Claude Code with Sonnet 5.5, Codex with GPT-5.4, or Cursor. Set a default per agent, or override per session. See Models for the full supported model list, BYOK details, and AWS Bedrock BYOK support.

Claude Code

Anthropic’s Claude models. Best balance of quality, reliability, and cost.Models: Opus 5.5, Opus 5, Fable 5.1, Fable 5, Sonnet 5.5, Haiku 5.5, Opus 4.8, Opus 4.7, Opus 4.6, Opus 4.5

Codex

OpenAI’s GPT models with configurable reasoning effort (low, medium, high, xhigh). GPT-6.1 Sol, GPT-6 Astra, GPT-6 Sol, GPT-5.6 Sol, and GPT-5.6 Terra also support max and ultra; GPT-6 Luna and GPT-5.6 Luna also support max.Models: GPT-6.1 Sol, GPT-6 Astra, GPT-6 Sol, GPT-6 Luna, GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna, GPT-5.5 Pro, GPT-5.5, GPT-5.4, GPT-5.3 Codex, GPT-5.4 Mini, GPT-5.4 Nano

Cursor

Multi-provider via Cursor CLI.Models: Composer 2.5, Composer 2, Composer 2 Fast, Composer 1.5, Opus 5.5, Opus 5, Sonnet 5.5, Haiku 5.5, Opus 4.8, Opus 4.7, Opus 4.6, Opus 4.5, GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.5, GPT-5.3 Codex, GPT-5.3 Codex Spark, Gemini 3.1 Pro, Gemini 3.5 Flash, Grok 4.7

Opencode

Multi-provider support: Anthropic, OpenAI, and OSS models.Models: Opus 5.5, Opus 5, Fable 5.1, Fable 5, Sonnet 5.5, Haiku 5.5, Opus 4.8, Opus 4.7, Opus 4.6, Opus 4.5, GPT-6.1 Sol, GPT-6 Astra, GPT-6 Sol, GPT-6 Luna, GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna, GPT-5.5 Pro, GPT-5.5, GPT-5.4, GPT-5.3 Codex, GPT-5.4 Mini, GPT-5.4 Nano, Gemini 3.1 Pro, Gemini 3.5 Flash, Grok 4.7, GLM 5.3, GLM 5.3 Flash, Kimi K3, DeepSeek V4 Pro, DeepSeek V4.1 Flash, MiniMax M2.7, MiniMax M3

Amp

Claude-powered with automatic model selection.

A minimal, extensible terminal coding harness with multi-provider support.Models: Opus 5.5, Opus 5, Fable 5.1, Fable 5, Sonnet 5.5, Haiku 5.5, Opus 4.8, Opus 4.7, Opus 4.6, Opus 4.5, GPT-6.1 Sol, GPT-6 Astra, GPT-6 Sol, GPT-6 Luna, GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna, GPT-5.5 Pro, GPT-5.5, GPT-5.4, GPT-5.3 Codex, GPT-5.4 Mini, GPT-5.4 Nano, GPT-5.2, Grok 4.7, GLM 5.3, GLM 5.3 Flash, Kimi K3, DeepSeek V4 Pro, DeepSeek V4.1 Flash, MiniMax M2.7, MiniMax M3

A tiny, open, native coding agent with multi-provider model support.Models: Opus 5.5, Opus 5, Fable 5.1, Fable 5, Sonnet 5.5, Haiku 5.5, Opus 4.8, Opus 4.7, Opus 4.6, Opus 4.5, GPT-6.1 Sol, GPT-6 Astra, GPT-6 Sol, GPT-6 Luna, GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna, GPT-5.5 Pro, GPT-5.5, GPT-5.4, GPT-5.3 Codex, GPT-5.4 Mini, GPT-5.4 Nano, Gemini 3.1 Pro, Gemini 3.5 Flash, GLM 5.3, GLM 5.3 Flash, Kimi K3, DeepSeek V4 Pro, DeepSeek V4.1 Flash, MiniMax M2.7, MiniMax M3

Triggers

These are just examples to give you a rough idea. You decide how to set this up, there are no limitations.

Scheduled

Run agents at regular intervals.

Schedule	Example
Hourly	Monitor for CI failures, new Sentry errors
Daily	Generate changelogs, post standup summaries
Weekly	Clean up stale TODOs, security scans
Monthly	Dependency audits, documentation reviews

Event-driven

React to events from your integrations in real-time.

Integration	Events
GitHub	PR opened, PR merged, issue created, comment added
GitLab	Merge request opened, merged, issue created
Sentry	New error, error spike, regression detected
Linear	Issue created, issue updated, status changed
Slack	Message in channel, @tembo mention

Webhooks

Trigger an agent on demand from CI, internal tools, or any service that can send an HTTP request. Send a POST request with an API key to the agent’s runs endpoint.

Deployment	URL
Tembo API	`https://api.tembo.io/v1/agents/{agentId}/runs`
Self-hosted	`https://{deployment-origin}/api/public-api/v1/agents/{agentId}/runs`

The agent ID is the UUID at the end of the agent’s URL in the dashboard (https://app.tembo.io/<workspace>/agents/<AGENT_ID>). Agent keys (macros) aren’t accepted and return 400.

curl -X POST "https://api.tembo.io/v1/agents/AGENT_ID/runs" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"issue_url": "https://github.com/org/repo/issues/42"}'

curl -X POST "https://tembo.example.com/api/public-api/v1/agents/AGENT_ID/runs" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"issue_url": "https://github.com/org/repo/issues/42"}'

Tembo responds with 201 once the run is queued. The response confirms the run is queued, not that it finished:

{
  "jobId": "96ea1b5b-fc94-4c4a-9b16-0dba674e4533",
  "agentId": "10dc26a2-9e2e-4a25-9f1b-4791fa2edf9c",
  "status": "queued"
}

Payload

The request body is optional. When you send one, it must be a JSON object. Send your fields at the top level; you don’t need to wrap them. Tembo passes the whole object to the agent as event context, so your instructions can reference it directly. For example, you could write instructions like “Fix the issue at the URL provided in the event payload”. If you need to transform the payload or fetch extra data before the agent starts, use an agent post-trigger hook. The hook can read TRIGGER_PAYLOAD and print Markdown, text, or JSON to standard output; Tembo adds that output to the agent’s starting context.

Authenticate with a query parameter

Some services only let you configure a URL, not headers. In that case, pass your API key as the apiKey query parameter:

curl -X POST "https://api.tembo.io/v1/agents/AGENT_ID/runs?apiKey=YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"ticket_id": "SUP-1234"}'

URLs can end up in logs and webhook delivery histories. Use the Authorization header when the sending service supports it.

Check run status

List the agent’s runs to see their status and linked session, newest first:

curl "https://api.tembo.io/v1/agents/AGENT_ID/runs" \
  -H "Authorization: Bearer YOUR_API_KEY"

Each run includes status, sessionId, durationMs, and errorMessage. See List agent runs and Create an agent run for the full API reference.

Errors

Status	Cause
`400`	The body isn’t a JSON object, or the agent ID isn’t a UUID.
`401`	The API key is missing, invalid, revoked, or expired.
`404`	No agent with that ID exists in the API key’s organization.

Creating an agent

You can create agents in the dashboard or define scheduled agents in a repository.

From a template

Go to Agents in the dashboard
Click Templates
Pick a template and click Use template
Customize the instructions and enable

Browse all templates

From scratch

Click New Agent
Write your instructions
Add triggers (schedule and/or event-based)
Select agent harness
Instruct MCP servers or integrations
Optionally select a project in run options so runs start from that prepared environment
Save and enable

From a repository

Store an agent’s instructions, runtime settings, and recurring schedules with the code it works on. Tembo syncs each JSON file directly inside .tembo/agents/ and creates or updates the corresponding agent in your workspace.

Prerequisites

Connect your GitHub, GitLab, or Bitbucket account to Tembo.
Activate the repository under Active Repositories on the Integrations page.
Make sure you can merge changes into the repository’s default branch.

Create a scheduled agent

Create the agent directory

At the root of your repository, create .tembo/agents/.Agent definitions must be .json files directly inside this directory. Nested files and other file types are ignored.

Add an agent definition

Create a file such as .tembo/agents/weekly-changelog.json:

{
  "name": "Weekly changelog",
  "prompt": "Summarize the changes merged during the last week and open a pull request that updates the changelog.",
  "schedule": "0 9 * * 1"
}

name and prompt are required. The schedule above runs every Monday at 09:00 UTC.

Merge the definition

Commit the file and merge it into the repository’s default branch. Tembo creates the agent during the next repository sync. Future changes to the file update the same agent.

Verify the agent

Open Agents, select the agent, and confirm its instructions and next scheduled run.

Configure multiple schedules

Use schedules to run the same agent on more than one cron schedule. Entries can be cron strings or objects with a cron property.

{
  "name": "Repository maintenance",
  "prompt": "Review the repository for stale dependencies and maintenance work. Open a pull request for safe updates.",
  "schedules": [
    "0 9 * * 1",
    { "cron": "0 12 * * 5" }
  ]
}

This example runs at 09:00 every Monday and 12:00 every Friday. Repository-defined schedules use UTC.

Use either schedule or schedules in one file, not both. You can configure up to 50 schedules for an agent.

Configure the runtime

You can keep the agent’s execution settings in the same file:

{
  "name": "Weekly changelog",
  "prompt": "Summarize the changes merged during the last week and open a pull request that updates the changelog.",
  "model": "codex:gpt-6.1-sol",
  "agentOptions": {
    "reasoningLevel": "high"
  },
  "mcpServers": ["linear"],
  "sandboxSize": "large",
  "projectId": "tembo",
  "artifactType": "PullRequest",
  "schedules": [
    { "cron": "0 9 * * 1" }
  ]
}

Field	Description
`model`	Agent harness and model key. See Models for supported options. `agent` is accepted as an alias, but do not set both.
`agentOptions`	Harness-specific options, such as `reasoningLevel`.
`mcpServers`	MCP server names that the agent can use. Connect them before the agent runs.
`sandboxSize`	VM size: `nano`, `micro`, `small`, `medium`, `large`, `xl`, `xxl`, or `ultra`.
`projectId`	A project UUID, or `tembo` to use the base environment. See Projects.
`artifactType`	Expected result: `PullRequest`, `ToolCall`, `Plan`, `Response`, `File`, `Service`, or `Source`.
`schedule`	One cron expression.
`schedules`	Up to 50 cron strings or objects containing a `cron` property.

Omit optional runtime fields to use the existing agent settings. For a new agent, omitted fields use your workspace defaults.

How synchronization works

Tembo matches an agent by its name within the connected repository. An agent with the same name managed by another repository remains separate.

Changing the prompt or runtime fields updates the repository’s matching agent.
Changing the configured cron expressions adds and removes schedules to match the file.
Omitting both schedule and schedules leaves existing schedules unchanged.
Setting "schedules": [] removes all schedules from the agent.
Disabling the agent in Tembo also keeps newly synced schedules disabled.
Removing the JSON file does not delete the agent. Delete or archive the agent from Tembo if you no longer need it.
Changing name creates a different agent instead of renaming the existing one.

Troubleshooting

The agent does not appear

Confirm that the repository is active in Integrations, the file is valid JSON with a .json extension, and it is directly inside .tembo/agents/ on the default branch. Repository changes are processed asynchronously, so allow time for the next sync.

The file is skipped

Tembo skips that file if it contains invalid JSON, unknown fields, conflicting fields such as both schedule and schedules, an unsupported model or runtime option, an inaccessible project, or an invalid or impossible cron expression. Other valid agent files in the directory continue to sync.

A schedule runs at the wrong time

Repository-defined cron schedules use UTC. Convert your desired local time to UTC and account for daylight saving time when you choose the expression.

A removed file still has an agent

Removing a file stops source-controlled updates but does not delete or archive its existing agent. Delete or archive the agent from Agents.

Macros

You can trigger a specific agent on demand from anywhere you can mention @tembo by using its macro.

@tembo !<macro-name> <prompt>

Tembo routes the request to the agent with that macro and passes the rest of the message as the prompt. For example, this triggers the agent named review:

@tembo !review please review this PR

Beyond code

Agents aren’t limited to engineering work. With MCP servers, agents can work across your entire stack.

Example	Trigger	What it does
Daily ARR report	Daily	Pulls revenue data from Stripe and posts to Slack
Feature requests from calls	Daily	Scans call transcripts for feature requests and creates Linear tickets
Customer health check	Weekly	Summarizes support tickets, usage trends, and churn signals in Slack
Enrich CRM records	New deal created	Pulls company info and adds context to Attio or your CRM
Create CRM tickets from Slack	`@tembo` in Slack	Creates a ticket in your CRM enriched with context from other connected integrations

Any tool with an MCP server can be used as an agent. You can also do this on-demand from Slack without setting up an agent:

@tembo What's our ARR right now? Check Stripe.

@tembo Summarize the feature requests from this week's sales calls and create Linear tickets for the top 5.

Get Started

Features

Actions

Integrations

Other

Agent harness

Triggers

Scheduled

Event-driven

Webhooks

Payload

Authenticate with a query parameter

Check run status

Errors

Creating an agent

From a template

From scratch

From a repository

Prerequisites

Create a scheduled agent

Configure multiple schedules

Configure the runtime

How synchronization works

Troubleshooting

Macros

Beyond code

​Agent harness

​Triggers

​Scheduled

​Event-driven

​Webhooks

​Payload

​Authenticate with a query parameter

​Check run status

​Errors

​Creating an agent

​From a template

​From scratch

​From a repository

​Prerequisites

​Create a scheduled agent

​Configure multiple schedules

​Configure the runtime

​How synchronization works

​Troubleshooting

​Macros

​Beyond code

Agent harness

Triggers

Scheduled

Event-driven

Webhooks

Payload

Authenticate with a query parameter

Check run status

Errors

Creating an agent

From a template

From scratch

From a repository

Prerequisites

Create a scheduled agent

Configure multiple schedules

Configure the runtime

How synchronization works

Troubleshooting

Macros

Beyond code