Sinatra

Sinatra is a hosted autonomous coding agent built by Behroz Saadat, a software engineer who founded the company in 2026. Assign a Linear ticket or add a GitHub label, and Sinatra reads your codebase, writes the code, runs your tests, and opens a draft pull request — all without requiring a new tool, IDE plugin, or dashboard for your team.

What It Is

Sinatra sits inside the issue trackers your team already uses — Linear and GitHub — and acts like an autonomous teammate. It picks up a ticket in seconds after assignment, works through a structured pipeline (reading the repo, planning the change, drafting the diff, running tests, self-reviewing), and delivers a pull request with a written summary and self-review notes. Merging always stays with the human reviewer and your existing CI rules.

How the Workflow Works

The three-step flow is designed to require zero new habits for most of the team:

• Assign or label — Assign a Linear issue to Sinatra, or add the Sinatra label to a GitHub issue. Setup is a one-time install of the GitHub App and Linear connection per workspace.
• Agent runs — Sinatra reads the codebase, plans the change (optionally brainstorming vague tickets first), drafts the diff, runs the configured test command, and fixes failures up to twice before opening a draft PR.
• Review and ship — The PR opens as a draft with a written summary and Sinatra's own self-review notes. Reviewers can comment and Sinatra will respond in-thread, push fixes to the same branch, and re-request review.

The .sinatrarc config file, committed to the repo, controls the test command, model selection, reasoning effort, and whether planning and brainstorm phases are active.

Model Flexibility and Cost Transparency

Sinatra is explicitly designed to run on credentials you already own. You can connect an Anthropic or OpenRouter API key, or a Claude or Codex/ChatGPT subscription, and Sinatra uses your credentials at provider list price with no LLM markup. If no credentials are configured, Sinatra supplies managed model access (Anthropic on the paid plan). The engine, model, and credential used are posted directly into the ticket after each run so no run is a black box. Per-repository model pinning via .sinatrarc is supported, including separate models for planning and brainstorm phases.

Platform Integrations

Sinatra currently supports two integrations, both available now:

• GitHub — Triggered by adding the Sinatra label to an issue. Reads PRs, checks, comments, and review threads; writes branches, PRs, and in-thread replies. Zero config after the GitHub App is installed.
• Linear — Triggered by assigning a ticket to Sinatra. Updates the ticket as work progresses and posts the PR link when ready.

Slack and Jira are listed on the roadmap but have not shipped. Each task runs in an isolated, short-lived sandbox checked out from your repository; sandboxes are never shared between workspaces and are archived automatically within a day of going idle.

Security and Code Handling

The about page states that Sinatra does not copy repository checkouts into its database, does not use code or content to train its own models, and does not sell it. Sandbox traffic runs through an egress proxy: public hosts are reachable and audited, while private and cloud metadata addresses are blocked. Credentials are stored encrypted and used only for the assigned task. The workflow does not merge or deploy — branch protection and required reviews remain in effect.

Current Status

Sinatra launched in 2026, is built on Google Cloud, and is operated remotely by a small founding team. The free tier requires no credit card and allows up to 5 tasks per day with a user-supplied API key or subscription. The paid Monthly plan is a flat per-member fee covering hosted sandboxes and orchestration with no LLM markup. Enterprise is available as a custom annual contract for teams with procurement and security review requirements.