Qoder

Agents connect to outside tools through MCP over stdio, SSE, HTTP or WebSocket, and a marketplace of connectors covers Linear, Postman, Notion, Grafana, MongoDB, Semgrep and more, with enterprise allowlists of up to 100 rules. Cloud Agents take GitHub repositories as resources, push branches and open pull requests with the customer's token, accept custom tools that the client runs, and talk to other agents over A2A. Team chat channels include Slack, Microsoft Teams, DingTalk, Lark, WeChat and WhatsApp.

Sourcedocs.qoder.com/cli/mcp-servers.mdread 2026-11-08

Experts Mode splits a task among named roles: a lead agent plans, a full stack engineer implements, a QA agent runs tests and builds, a UI operator checks the result in a browser and a code reviewer reviews it. The command line tool runs subagents and headless jobs in CI. QoderWake groups Wakers under a leader and runs multistage work through WakerFlow, and Cloud Agents deployments run saved agent configurations on a schedule or on demand.

Sourcedocs.qoder.com/user-guide/quest/experts-mode.mdread 2026-11-08

Each repository with fewer than 10,000 files is indexed automatically, with only vector data kept in the cloud and exclusions set in a .qoderignore file. Repo Wiki generates documentation pages and knowledge cards from the code, exports them into the repository, shares them through Git and updates them automatically as the code changes. A knowledge base holds uploaded files at user or workspace scope for the agent to search while it works.

Sourcedocs.qoder.com/qoder/indexing.mdread 2026-11-08

By default the desktop app asks before every action, with auto approve and full access as opt in modes that admins can switch off. The command line tool adds allow, ask and deny rules and lets admins disable its bypass mode, and Cloud Agents set each tool to always allow, always ask or always deny. Work runs in a sandbox that can write only to the workspace, plans can be approved from a phone, and changes come back as a diff to review before commit.

Sourcedocs.qoder.com/qoder/approval-and-sandbox.mdread 2026-11-08

Teams and Enterprise accounts sign in through SAML 2.0 or OIDC with automatic provisioning by email domain, and roles cover Admin, Config Admin, Member and Directory Member, with admin control over models and MCP servers. Qoder holds SOC 2 Type II for security, availability and confidentiality and states ISO 27001 certification. Users can opt out of sharing data for product improvement, depending on plan, and data with identifiers removed is kept for up to five years.

Sourcedocs.qoder.com/account/teams/sso.mdread 2026-11-08

Cloud Agents keep a full event history for each session, streamed live, including the credits spent on every model call, and QoderWake keeps a record of every run. Teams and Enterprise analytics show the share of committed code written with AI, retained lines and completion acceptance. Enterprise adds an audit log of administrative actions, which records configuration changes and leaves out user content.

Sourcedocs.qoder.com/cloud-agents/api/sessions/events/overview.mdread 2026-11-08

Global and project memories persist across sessions and can be viewed and cleared in settings, with low scoring memories pruned over time. The command line tool reads AGENTS.md at user, project and local level and can save memories automatically when switched on. Cloud Agents keep memory stores that are versioned, can be redacted and last until deleted, and Wakers keep personal, project, group and member memory in an editable file whose earlier versions can be restored.

Sourcedocs.qoder.com/cloud-agents/memory-stores.mdread 2026-11-08

Data sits on servers in the United States, Singapore and Japan, with no region picker, and a separate service runs for mainland China. Teams that need execution on their own machines can run Cloud Agents in a self hosted environment that uses the customer's own workers, and QoderWake can be installed on the customer's own Linux server, so the agent's work runs inside infrastructure the customer controls.

Sourcedocs.qoder.com/cloud-agents/environments.mdread 2026-11-08

A marketplace sells skills, plugins and connectors, with more than 20,000 skills and over 100 plugins on offer, and enterprises can curate private collections of them. QoderWake comes with preset Wakers for release, development, review and testing work alongside custom ones, and Experts Mode supplies five ready roles. Each preset does a job on its own, so a team can adopt one without taking the others.

Sourceqoder.com/marketplaceread 2026-11-08

Background work starts without anyone at the keyboard. QoderWake tasks fire on a schedule, on repository issue and pull request events, on an API call authenticated with a personal token, or by polling, with up to five triggers per task, and Wakers answer @mentions in chat. Cloud Agents deployments run on a cron schedule, desktop Automations run scheduled jobs locally or in the cloud, and outbound webhooks report session events back to the customer's own systems.

Sourcedocs.qoder.com/qoderwake/automated-tasks.mdread 2026-11-08

Users pick a model or a tier: Auto, which routes each request and costs about half the credits, Ultimate, Fast or Efficient. Named models include Qwen3.8 Max and Flash, DeepSeek V4 Pro and Flash, GLM 5.3, Kimi K3.1 Pro and MiniMax M3.1 Flash. Custom models connect through OpenAI, Google, OpenRouter, DeepSeek, Kimi or any OpenAI or Anthropic compatible endpoint, billed by that provider, and enterprise admins decide which model families teams may use.

Sourcedocs.qoder.com/qoder/model-selector.mdread 2026-11-08

Cloud Agents expose a REST API at api.qoder.com, labeled beta, with bearer authentication by personal token or by short lived service account tokens, and SDKs for Python, TypeScript, Go and Java. A separate Agent SDK for TypeScript, Python and Java embeds Qoder's agent in other software, the command line tool runs headless in CI, and Cloud Agents also speak the A2A protocol and offer an MCP connector of their own.

Sourcedocs.qoder.com/cloud-agents/api/conventions/overview.mdread 2026-11-08

Cloud Agents can grade their own results: an Outcomes setting sends the work to a separate evaluator that scores it against a rubric and sends the agent back for another pass, for up to 20 rounds and three by default. In the IDE, an ultra review command reviews changes before commit, security scans run at three depths, and Experts Mode's QA agent runs the project's tests and builds as part of each task.

Sourcedocs.qoder.com/cloud-agents/api/sessions/outcomes.mdread 2026-11-08

The agent drives a built in browser with live preview of local apps, and Computer Use controls the desktop on macOS 14 and Windows 10 or later, operates an external Chrome or Edge browser through an extension, and can record a sequence of steps and replay it. In Experts Mode a UI operator checks finished work in the browser, and Cloud Agents offer browser use, labeled beta, with a live view.

Sourcedocs.qoder.com/qoder/computer-use.mdread 2026-11-08