llms.txt
# Factory
> Factory is the agent-native software factory: a system for delivering production software with AI agents (Droids) that plan, build, review, and ship under human direction. Factory ships the Droid CLI, the Factory desktop and web apps, Droid Computers, Missions, and an enterprise platform used by financial services, defense, healthcare, telecom, and technology teams.
Primary product: [Factory Droid](https://factory.com) - autonomous coding agent available via CLI, web, desktop, Slack, Jira, and Linear surfaces.
## When to use Factory
Use Factory for software engineering work with a clear outcome that can be verified in a repository, issue tracker, terminal, or CI pipeline. Best-fit jobs include implementing scoped features, fixing bugs, reviewing pull requests, generating and running tests, investigating incidents, maintaining documentation, and automating recurring delivery workflows.
Choose an execution surface based on the job:
- Use the Droid CLI for interactive engineering work in a local repository or headless execution in CI.
- Use Missions for long-running, multi-phase work that needs planning, worker coordination, validation, and human checkpoints.
- Use Automations for scheduled or event-driven work triggered by systems such as Slack or GitHub.
- Use the Factory Public API when software needs typed programmatic access to sessions, computers, automations, organization usage, service accounts, or wikis.
## How agents should call Factory
- Read [the documentation index](https://docs.factory.com/llms.txt) first, then fetch the task-specific `.md` page it links to.
- Discover API operations from [the OpenAPI specification](https://factory.com/openapi.json). Every operation has a unique `operationId`, typed parameters and responses, and a description.
- Authenticate Public API requests with `Authorization: Bearer $FACTORY_API_KEY`. Create keys in [Factory API Keys settings](https://app.factory.ai/settings/api-keys), and store them securely before making requests.
- Send `Accept: text/markdown` when retrieving the Factory homepage or an unknown path for a compact Markdown response.
## Website API errors
The Factory Public API runs on api.factory.ai. On the marketing website, unknown /api paths and partner application errors return RFC 9457 application/problem+json with code and hint extensions. Partner errors retain the existing error field.
POST /api/partners reports its actual per-instance, per-IP limit in RateLimit and RateLimit-Policy structured fields (IETF RateLimit draft): five requests per ten minutes. Local 429 responses include Retry-After in seconds; valid upstream Retry-After values are preserved when supplied. Honor Retry-After before retrying, or use exponential backoff when absent. These counters do not describe an API-wide quota.
## Articles
> Long-form guides and comparisons on software factories, AI coding agents, and the Droid platform. Articles are not linked from the main navigation but are individually indexed.
- [Droid vs Devin for enterprise engineering workflows](https://factory.com/articles/droid-vs-devin-enterprise-workflows): Compare Droid vs Devin for interactive coding, delegated multi-step work, and review handoffs. Separate saved instructions, persistent compute, and model configuration.
- [Factory vs Cognition for enterprise agent controls](https://factory.com/articles/factory-vs-cognition-enterprise-controls): Compare Factory vs Cognition on administering Droid and Devin: user access, model policy, command permissions, automation ownership, and audit records.
- [Factory vs Devin: where your engineering data goes](https://factory.com/articles/factory-vs-devin-privacy-deployment): Compare Factory vs Devin on where code runs, which services receive model requests, and what sessions and telemetry retain. Understand the limits of private deployment.
- [Backup recovery tests for agent-generated services](https://factory.com/articles/backup-recovery-testing-coding-agents): Backup recovery testing for coding agent changes verifies restore procedures, integrity, dependencies, and recovery objectives with reviewable evidence.
- [CSP testing for coding agent web changes](https://factory.com/articles/content-security-policy-testing-coding-agents): Content Security Policy testing for coding agent changes verifies directives, nonces, report-only rollout, and browser behavior without weakening protection.
- [CORS checks for agent-generated web services](https://factory.com/articles/cors-testing-agent-generated-services): CORS testing for agent-generated services verifies allowed origins, preflight handling, credentials, caching, and failure behavior in browsers.
- [A model-agnostic alternative to OpenAI dots](https://factory.com/articles/model-agnostic-openai-dots-alternative): Compare OpenAI dots with Factory on model selection, pricing and usage limits, cloud computers, projects, background work, and permissions.
- [The multi-agent architecture behind reliable missions](https://factory.com/articles/multi-agent-architecture-missions): Luke Alvoeiro explains the multi-agent architecture behind Factory Missions. Explore planning, independent validation, structured handoffs, and selective parallelism.
- [OpenAI dots vs Factory for software development](https://factory.com/articles/openai-dots-vs-factory): Compare OpenAI dots with Factory on software delivery, cloud computers, background work, model selection, pricing, and action approvals.
- [Recursive self-improvement needs a better measure](https://factory.com/articles/recursive-self-improvement-matan-interview): Matan Grinberg discusses recursive self-improvement and AI adoption with Sabrina Halper. Separate model improvement, engineering feedback, token spending, and useful outcomes.
- [Session cookie tests for coding agent changes](https://factory.com/articles/session-cookie-testing-coding-agents): Session cookie testing for coding agent changes verifies scope, transport protections, expiration, rotation, and logout behavior across application routes.
- [Tenant isolation tests for coding agent changes](https://factory.com/articles/tenant-isolation-testing-coding-agents): Tenant isolation testing for coding agent changes verifies authorization, storage boundaries, cache keys, and background jobs before shared services ship.
- [What are OpenAI dots, and what are the alternatives?](https://factory.com/articles/what-are-openai-dots): What are OpenAI dots? Explore their cloud computers, ongoing work, models, and approvals alongside Factory, a model-agnostic alternative for software development.
- [API rate limit tests for agent-generated services](https://factory.com/articles/api-rate-limit-testing-coding-agents): API rate limit testing for coding agent changes verifies quotas, 429 responses, retry guidance, and recovery before a service reaches production.
- [File upload security tests with coding agents](https://factory.com/articles/file-upload-security-testing-coding-agents): File upload security testing with coding agents verifies type, size, storage, authorization, and retrieval controls with reproducible fixtures.
- [Memory leak tests for agent-generated services](https://factory.com/articles/memory-leak-testing-agent-generated-code): Memory leak testing for agent-generated code measures retained memory across repeated workloads and gives reviewers reproducible evidence.
- [SSRF testing for coding agent network changes](https://factory.com/articles/ssrf-testing-coding-agent-changes): SSRF testing for coding agent changes verifies URL validation, network boundaries, redirects, and DNS behavior before server-side requests ship.
- [Telemetry schema checks for coding agent changes](https://factory.com/articles/telemetry-schema-checks-coding-agents): Telemetry schema checks for coding agent changes keep metric, span, and log attributes stable, bounded, and safe for downstream systems.
- [Authorization tests for coding agent changes](https://factory.com/articles/authorization-testing-coding-agent-changes): Authorization tests for coding agent changes require denied-path coverage, object-level cases, policy fixtures, and evidence at the enforcement point.
- [Concurrency testing for coding agent changes](https://factory.com/articles/concurrency-testing-coding-agent-changes): Concurrency testing for coding agent changes combines race detectors, controlled execution orders, invariants, and stress runs with reproducible evidence.
- [Queue consumer safety in agent-generated code](https://factory.com/articles/queue-consumer-safety-coding-agents): Queue consumer safety in agent-generated code depends on idempotent effects, correct acknowledgments, bounded retries, and tests for duplicate delivery.
- [Time zone testing for agent-generated code](https://factory.com/articles/time-zone-testing-agent-generated-code): Time zone testing for agent-generated code covers daylight-saving transitions, database updates, ambiguous local times, and deterministic clocks.
- [Webhook verification for coding agent changes](https://factory.com/articles/webhook-verification-coding-agents): Webhook verification for coding agent changes requires raw-body signature checks, replay defenses, secret rotation, and negative tests before merge.
- [Database query plan review for agent changes](https://factory.com/articles/database-query-plan-review-coding-agents): Database query plan review for coding agent changes compares representative plans, measured execution, and schema context before a performance-sensitive merge.
- [Kubernetes manifest validation before rollout](https://factory.com/articles/kubernetes-manifest-validation-coding-agents): Kubernetes manifest validation for coding agent changes should use the target API server, show field ownership, and keep rollout decisions with reviewers.
- [Log redaction for agent-generated changes](https://factory.com/articles/log-redaction-coding-agents): Log redaction review for coding agent changes starts with a data inventory, tests structured output, and verifies that useful diagnostics survive.
- [Protobuf compatibility in agent-generated changes](https://factory.com/articles/protobuf-compatibility-coding-agents): Protocol Buffer compatibility review for coding agent changes checks field history, generated clients, mixed-version behavior, and rollback safety.
- [Retry safety for agent-generated application code](https://factory.com/articles/retry-safety-agent-generated-code): Retry safety review for agent-generated code verifies idempotency, retry ownership, backoff, time limits, and evidence from failure-path tests.
- [API client generation with coding agents](https://factory.com/articles/api-client-generation-coding-agents): API client generation with coding agents should start from a reviewed contract, use a pinned generator, and validate compatibility before updating consumers.
- [Build provenance for coding agent artifacts](https://factory.com/articles/build-provenance-coding-agent-artifacts): Build provenance connects coding agent artifacts to source, workflow, and inputs. Generate attestations in CI and verify them before release.
- [Code ownership for coding agent pull requests](https://factory.com/articles/code-ownership-coding-agent-pull-requests): Code ownership routes coding agent pull requests to people who understand the affected systems. Keep ownership rules current, specific, and protected.
- [Container image hardening with coding agents](https://factory.com/articles/container-image-hardening-coding-agents): Container image hardening with coding agents works best when the task fixes a measured image risk, preserves reproducibility, and produces evidence for review.
- [Keeping test data private with coding agents](https://factory.com/articles/test-data-privacy-coding-agents): Test data privacy for coding agents requires approved fixtures, restricted access, and checks that prevent sensitive records from entering prompts, logs, or commits.
- [Accessibility testing with coding agents](https://factory.com/articles/accessibility-testing-coding-agents): Accessibility testing with coding agents combines deterministic scans, keyboard flows, semantic inspection, and human review. Build useful evidence into every UI change.
- [Agent readiness: fix the repos slowing your rollout](https://factory.com/articles/agent-readiness-rollout): Use agent readiness to find the repositories blocking an AI rollout, prioritize fixes, and measure progress with Factory’s reports and dashboard.
- [Agent readiness can cut repeated token work](https://factory.com/articles/agent-readiness-token-efficiency): Agent readiness exposes avoidable setup, search, and validation work. Connect repository improvements to token efficiency without inventing a savings benchmark.
- [How to write an AGENTS.md file](https://factory.com/articles/agents-md-guide): Write an AGENTS.md file with verified commands, clear scope boundaries, and useful completion checks. Test the instructions on a small task before expanding them.
- [AI agent controls for private development networks](https://factory.com/articles/ai-agent-controls-private-networks): AI agent controls for private networks need enforceable permissions, sandboxing, managed model policy, and review gates. Build a testable policy for Droid.
- [Protecting API compatibility with coding agents](https://factory.com/articles/api-compatibility-coding-agents): API compatibility checks help coding agents change services without surprising consumers. Compare contracts, trace callers, and validate old and new behavior.
- [Model routing has to account for the warm cache](https://factory.com/articles/cache-aware-model-routing): Model routing can lose its price advantage when a switch discards cached context. Factory’s research explains why task state belongs in the decision.
- [Audit trails for coding agent changes](https://factory.com/articles/coding-agent-audit-trails): A coding agent audit trail connects identity, intent, tool activity, validation, review, and deployment. Learn what evidence to retain without collecting unnecessary data.
- [Configuration drift repair with coding agents](https://factory.com/articles/coding-agent-configuration-drift): Configuration drift repair with coding agents starts from approved baselines, separates evidence from fixes, and keeps changes reviewable.
- [Context compaction must preserve the next step](https://factory.com/articles/coding-agent-context-compaction): Context compaction should preserve decisions and file state, not just shrink a transcript. Factory’s published study shows what to test after a long coding session.
- [Safer dependency updates with coding agents](https://factory.com/articles/coding-agent-dependency-updates): AI dependency updates need more than a version bump. Scope compatibility work, inspect transitive changes, run focused tests, and keep every update reviewable.
- [Safer database migrations with coding agents](https://factory.com/articles/coding-agent-database-migrations): Database migrations with coding agents need staged schema changes, compatibility tests, rollback evidence, and human approval without hiding operational risk.
- [Removing stale feature flags with coding agents](https://factory.com/articles/coding-agent-feature-flag-cleanup): Feature flag cleanup requires proof that a flag is stale. Coding agents can remove the obsolete branch and evaluation call while testing the permanent behavior.
- [Localization testing with coding agents](https://factory.com/articles/coding-agent-localization-testing): Localization testing with coding agents checks translated interfaces for layout, formatting, direction, missing messages, and functional regressions.
- [Coding agent observability that reviewers can verify](https://factory.com/articles/coding-agent-observability-changes): Coding agent observability work needs stable signal names, bounded cardinality, and behavioral evidence. Learn how to review instrumentation changes.
- [Coding agent productivity metrics that matter](https://factory.com/articles/coding-agent-productivity-metrics): Coding agent productivity metrics should connect cost and usage to delivery, quality, and business priorities. Measure outcomes without rewarding code volume.
- [Runtime upgrades with coding agents](https://factory.com/articles/coding-agent-runtime-upgrades): Runtime upgrades with coding agents can turn compatibility discovery, dependency updates, and validation into a reviewable migration.
- [Coding agent task specifications that work](https://factory.com/articles/coding-agent-task-specifications): A coding agent task specification needs a bounded outcome, repository context, acceptance criteria, and verification steps. Use this practical structure for delegated work.
- [Test fixture maintenance with coding agents](https://factory.com/articles/coding-agent-test-fixture-maintenance): Test fixture maintenance with coding agents can remove hidden dependencies, reduce setup cost, and preserve the behavior each test proves.
- [Visual regression testing with coding agents](https://factory.com/articles/coding-agent-visual-regression-testing): Visual regression testing with coding agents combines stable screenshots, targeted interaction, and human review of intentional UI changes.
- [Coding agents in monorepos, with clear boundaries](https://factory.com/articles/coding-agents-in-monorepos): AI coding agents in monorepos need explicit package boundaries, local instructions, dependency graphs, and affected tests. Make large repositories easier to change safely.
- [Cross-repository coding agent coordination](https://factory.com/articles/cross-repository-coding-agents): Cross-repository coding agents need explicit dependency order, isolated workspaces, compatible changes, and release evidence. Use this plan for coordinated updates.
- [What enterprise software teams build with Factory](https://factory.com/articles/enterprise-software-customer-projects): Published Factory case studies show enterprise software work across modernization, secure packages, financial products, search infrastructure, and day-0 launches.
- [Enterprise software modernization beyond the demo](https://factory.com/articles/enterprise-software-modernization): Enterprise software modernization needs clear scope and reviewable evidence. Comarch’s published migrations show how Factory Missions fit that work.
- [EU AI deployment beyond the regional endpoint](https://factory.com/articles/eu-sovereign-ai-deployment): EU AI deployment requires separate decisions about session content, inference, identity, and control-plane ownership. Compare regional Factory hosting with Factory Private.
- [How to evaluate coding agents on your codebase](https://factory.com/articles/evaluating-coding-agents-on-your-codebase): Evaluate coding agents with representative tasks, reproducible environments, independent acceptance checks, and a record of cost and human intervention.
- [Evaluating self-improving software after release](https://factory.com/articles/evaluating-self-improving-software): Evaluating self-improving software requires stable checks and comparable work. Connect workflow changes to accepted outcomes, human correction, cost, and later failures.
- [Factory and Claude Code: compare completed work](https://factory.com/articles/factory-claude-code-token-efficiency): Compare Factory and Claude Code token efficiency using accepted changes, billed usage, cache behavior, and reruns. Keep model and harness effects separate.
- [Factory Managed vs Private: choosing the boundary](https://factory.com/articles/factory-managed-vs-private): Compare Factory Managed vs Private across control-plane ownership, model access, telemetry, and operations. Choose a deployment around your actual requirements.
- [Factory Private architecture inside your boundary](https://factory.com/articles/factory-private-architecture): Factory Private architecture puts the control plane in you
