Warp

Warp connects to Slack, Linear and GitHub so an agent can be summoned by mentioning @Warp inside those tools to investigate bugs, suggest fixes or open pull requests, and Factories adds Teams, Jira, GitLab and source forge integration plus MCP server connection management; the published Agent Kits show tool calling across CRM, billing, support, analytics, documentation, monitoring and search classes including HubSpot, Stripe, Intercom, Metabase, BigQuery, Notion, Mintlify, Sentry, Apollo, Grain, Exa and Octolens, and agents run arbitrary CLI tooling inside their sandbox.

Sourcewarp.dev/pricing integrations section, warp.dev/factories and warp.dev/agent-kitsread 2026-08-30

Multi-agent orchestration on the generally available Automation Platform lets a parent agent spawn and coordinate child agents across local and cloud runs, for supervisor and worker, fan-out, critic, DAG and swarm workflows, started from the Warp app, the Oz CLI, the Oz web app or the Oz API and SDK, with parent and child conversations and artifacts inspectable; local and cloud agents hand work off to each other. Warp Factories, in early access, adds a six stage factory line defined in factory.yaml.

Sourcedocs.warp.dev platform orchestrationread 2026-09-29

Codebase Context semantically indexes a repository's Git-tracked files so agents can retrieve relevant code, and it applies the same way in the app, the CLI and cloud runs; paid plans carry the highest indexing limits and Enterprise adds custom codebase indexing. Warp Drive stores team workflows, prompts and rules, and MCP servers connect external sources per agent.

Sourcedocs.warp.dev agents capabilities (Codebase Context) and warp.dev/pricingread 2026-09-29

The vendor states customers control exactly how much autonomy agents have, from approving every step to allowing full autonomy, configurable at individual and enterprise level; in Factories the line stops wherever the customer decides with named gates for spec approval, code review and QA, the vendor states merge is always yours, and the platform documents approval gate enforcement before merge, checkpoint pause and resume state, fine grained permissions per agent, network egress policy enforcement and centrally managed permissions with centrally configured agent access.

Sourcewarp.dev/pricing privacy section, warp.dev/factories and warp.dev/enterpriseread 2026-08-30

SOC 2 Type 2 attestation obtained from an accredited third party with the report requestable through a Trust Center, SAML based SSO on Business and enforced single sign on through a provider such as Okta plus domain verification on Enterprise, AES 256 or higher at rest and TLS 1.3 in transit, Zero Data Retention with all contracted LLM providers so no customer data is retained, stored or used for training, admin enforceable telemetry, AI and secret redaction controls, and in Factories a non human identity per agent with fine grained permissions, secrets storage and rotation, credential handling and network egress policy enforcement.

Sourcewarp.dev/legal/security and warp.dev/pricingread 2026-08-30

Every generally available cloud agent run keeps a persistent record with status, metadata, cost and a session transcript covering the prompt, plan, commands, logs, files changed, outputs and follow-up messages; the Runs page in the Oz web app finds runs by source, status, trigger or owner, opens the transcript and shares it for review, and the control plane records run history. Enterprise adds an Analytics API and the client offers a Network Log.

Sourcedocs.warp.dev platform (Viewing cloud agent runs, Platform architecture) and warp.dev/pricingread 2026-09-29

Agent Memory is a persistent store on Warp that agents read from and write to across harnesses (Warp Agent, Claude Code, Codex), attachable to agents read-only or read-write, so facts, decisions and outcomes carry from one conversation to the next. Shipped today: cloud conversation storage, Warp Drive objects and run continuity. Agent Memory is in research preview, enabled per team for design partners through a waitlist.

Sourcedocs.warp.dev agents (Agent Memory, research preview) and warp.dev/pricingread 2026-09-29

Enterprise plans document self hosted cloud agents on the customer's own infrastructure and inference routed through the customer's own cloud, Factories offers Warp's cloud or fully self hosted inside the customer's own VPC under their existing retention and compliance rules with self hosted workers and the customer's own code forge for factory config, compute is Docker or Kubernetes backed with per run sandbox provisioning, and the client itself is a native application for macOS, Windows and Linux whose source is public under MIT and AGPL. The hosted SaaS path stores Warp Drive and account data on Google Cloud Platform with database locations in the United States and no region selection offered.

Sourcewarp.dev/pricing, warp.dev/factories and warp.dev/legal/securityread 2026-08-30

Agent Kits is a published gallery of pre built agent packages described as one click to deploy with full control to customize, cataloged across Growth, Sales, Recruiting, Research, Risk Trust and Safety, and Engineering, each shipping a named trigger, a skill file, a default model and its integrations, including Conversion Ratio Optimization, Churn Intervention, Documentation, Social Monitoring, Go to Market, Account Status, POC, Recruiting Sourcing, Competitive Intelligence, Fraud Detection and Client Release agents; Warp Drive additionally stores reusable workflows, prompts and rules as unlimited objects on paid plans, and factory.yaml defines reusable agent types such as FOREMAN and REVIEW as code.

Sourcewarp.dev/agent-kits and warp.dev/factoriesread 2026-08-30

On the generally available Automation Platform a trigger is anything that starts a cloud agent without anyone typing a prompt: Scheduled Agents run on a cron schedule (oz schedule create), GitHub Actions and other CI events start runs, Slack, Linear and Jira integrations pick up work, and the Oz API starts runs programmatically. The terminal, CLI, web app and IDE surfaces cover invoked runs. Warp Factories, in early access, adds further webhooks.

Sourcedocs.warp.dev platform triggers (Triggers overview, Scheduled Agents, GitHub Actions)read 2026-09-29

A model picker exposes Fable 5, Claude Opus 5, GPT-5.6, Gemini 3.1 Pro, GLM 5.2, Kimi K3 and others with an Auto option, and the customer can run different tasks on different models in parallel or switch model mid task; Business adds bring your own API keys and custom inference endpoints, Enterprise routes inference through the customer's own cloud as BYOLLM at no credit cost, Factories sets a default model per factory and overrides it per agent in factory.yaml giving model routing per pipeline stage, and the vendor states customers may bring their own model or harness rather than using Warp's.

Sourcewarp.dev/pricing models section and warp.dev/factoriesread 2026-08-30

A REST API at app.warp.dev/api/v1/agent/runs accepts a prompt with bearer token authentication and returns a run id and state, a TypeScript SDK exposes client.agent.run, a CLI runs warp agent run-cloud against a named environment, and a warp-factory MCP server accepts tool calls such as send_task against a factory uid; Warp also consumes MCP servers with connection management, ships webhook signature verification, a settings file allowing users and agents to configure Warp programmatically and sync setups across environments, an Enterprise Analytics API, and a client whose source is public on GitHub under MIT and AGPL with docs published as an open repository.

Sourcewarp.dev/factories and docs.warp.devread 2026-08-30

Evaluation lives inside Warp Factories, which is in closed early access by application: evals run on the customer's own work with a pass rate tracked over time, benchmarks comparing models against each other on the customer's tasks with relative score and cost per run, benchmark regression detection, an evals and scoring pipeline where every run is scored, self improvement loops in which observer agents open pull requests against the factory config, and insights including cost per pull request and automation rate. No evaluation, benchmarking or regression surface is documented for the generally available terminal, Agent CLI or cloud agents.

Sourcewarp.dev/factories quality loop sectionread 2026-08-30

Computer Use is documented for generally available cloud agents: in a sandboxed cloud desktop the agent takes screenshots, records annotated video, clicks, drags, types and uses keyboard shortcuts to test UI changes, enabled per run with --computer-use in the Oz CLI, the computer_use_enabled field in the Oz API (default on for Warp's built-in environments) or a toggle in the Oz web app, with screenshots and videos attached to pull requests. The desktop ships with the platform.

Sourcedocs.warp.dev agents capabilities (Computer Use)read 2026-09-29