cayu-dev/cayu

Find public concepts · Run the example

Cayu is an open-source Python framework for building and running domain-specific long-horizon agents, with a durable runtime built in.

Build your agent's harness from Python components and defaults: model providers, context, tools, integrations, knowledge, memory, policies, domain logic and verification. Choose the capabilities your job needs, then evaluate and operate the resulting agent on your infrastructure or with optional Cayu Cloud.

The integrated runtime carries that work through model and tool calls, durable sessions, task dispatch, leased workers, resumable workflow steps, approvals and recovery. Applications can use those execution capabilities directly without a separate workflow engine.

Applications own their user interface, authentication, business authorization, domain rules and completion criteria. See the public concepts and source map for the components to build with.

We built Cayu while developing domain-specific agents that needed to carry consequential work through model calls, tools, human decisions and failures. Those applications needed both a configurable harness and dependable execution.

We began by building agents with SDKs and frameworks including the Claude Agent SDK, Mastra, and LangGraph. They helped us implement the model-and-tool loop quickly. Production quality required deeper control of the loop itself: context assembly, model and tool invocation, output validation, and failure handling.

Production quality also depended on everything around that loop: scheduling, durable state, credentials, execution environments, human intervention, recovery, cost attribution, and evaluation. Cayu gives applications control of that entire agent execution lifecycle.

Agent prototypes are easy to start. Production failures happen at the boundaries:

• a process dies after a side effect but before state is recorded;
• a model requests a valid tool with the wrong authority;
• a run needs human input or approval halfway through;
• context grows until a provider rejects the next request;
• retries, forks, or subagents lose cost and causal attribution;
• operators cannot reconstruct what happened from prompt text alone; or
• evals test final prose while missing the runtime trajectory.

Cayu treats these as runtime contracts. Important actions become structured events; tool authority and recovery are explicit; configured durable stores let transcripts and checkpoints survive process boundaries; and the same public seams support local development, tests, control-plane inspection, and hosted deployments.

The same framework provides the execution contracts behind those capabilities:

• durable sessions, transcripts, events, checkpoints, resume and interruption;
• controlled tool effects, policies, approvals and idempotency identities;
• human-input checkpoints, approval resolution and manual recovery;
• task dispatch, worker ownership, resumable workflows and child-session lineage;
• usage accounting, run limits, budgets, pricing and causal attribution; and
• execution admission, recovery and reconciliation of uncertain external effects.

Configure the stores, adapters, policies and execution boundaries for your job. Native Python tools are trusted application code; configured isolation and external receipts determine the guarantees at the application boundary.

Want executable evidence before adopting Cayu? The runtime qualification guide maps tool roundtrips, durable continuation, approval gating, and reviewed-proposal binding to existing tests. It provides a pinned installed-wheel setup, a small credential-free acceptance plan, the restartable support integration, and broader release qualification, with report interpretation and coverage limits.

Need a non-coding agent that survives restarts and waits for approval? Follow the restartable support-agent example. It investigates an order, asks for clarification, proposes a replacement, and waits for verified representative approval before executing it. The guide and executable ship with Cayu:

cayu guide order-support
python -m cayu.examples.order_support --help

For a model-only agent, start with the generated project below.

The generated project is the recommended path for both humans and coding agents. Cayu requires Python 3.11 or newer.

You can give a coding agent one request: “Run pip install cayu and create a code review agent.”

pip install cayu pytest
cayu new myagent
cd myagent

cayu inspect --json
cayu check --fail-on warning --json
pytest
cayu eval run

# After configuring the provider through configuration/settings.py or CAYU_PROVIDER:
python run.py --message "Review this change."

For a portable operator snapshot that performs no agent, tool, recovery, repair, or live-provider work, run cayu doctor --bundle cayu-support.zip. The command reuses maintained-service diagnostics when selected and forces CayuApp and project control-plane stores backed by the built-in SQLite and PostgreSQL implementations into diagnostic read-only validation without creating or migrating them. Session history is excluded unless you explicitly repeat --session. See Diagnostic support bundles for the contents, redaction boundary, hard limits, and exit codes.

The scaffold is credential-free and includes:

• a composition-only app.py with a process-scoped build_app() factory;
• one model-only agent with no required tools;
• complete tracked homes for prompts, tools, policies, environments, workflows, operations, knowledge, memory, integrations, observability, and domain code;
• explicit provider, storage, runtime, and agent-registration seams;
• a hermetic runtime test and output eval; and
• a source-controlled [tool.cayu.scaffold] plan plus AGENTS.md and the minimal CLAUDE.md bridge.

Open the generated project, describe the requested job in the existing agent, and keep its public test/eval seam intact.

For an explicit, editable repository-coding starter with bounded file and Git tools, durable knowledge, background review delegation, and human input, opt in to the maintained composition:

cayu new mycoder --preset coding
cd mycoder
cayu check --fail-on warning --json
pytest -q tests/test_coding_composition.py
python run.py --agent mycoder --message "Implement the requested change."

Add --execution docker for the maintained GitHub-independent coding product. Its build_coding_product_application() front door produces durable patch_ready_for_delivery evidence after required checks, source copy-back, Git inspection, and configured review gates settle. It performs no commit, push, pull-request, CI, or merge effect. See the maintained coding product.

Add --with remote-git-delivery to generate the optional host-side broker seam. It prepares the exact patch-ready tree for application approval, then can create and push one exact commit to a new configured branch without exposing remote or credential authority to the coding agent. See approved remote Git delivery.

Add --with github-delivery to include the subsequent host-only PR, exact-head checks, review, and provenance-labelled follow-up seam. It implies remote Git delivery and does not authorize merge. See GitHub delivery.

The generated repository starts from a clean Git commit, and each new product run requires a clean committed Git source baseline. It also requires git, rg, and the POSIX descriptor-relative filesystem primitives used by secure LocalWorkspace path operations. Unsupported hosts fail during generation or application construction. Its local workspace and runner are trusted-host development adapters, not a hostile-code sandbox. The default scaffold is the complete Cayu application convention. Select --preset agent, --preset service, or --preset coding; optional capabilities share the same application layout.

Discover or review the Rails-style generation plan before writing:

cayu new --list-presets --json
cayu new --list-capabilities --json
cayu new myagent --preset agent --provider anthropic --dry-run --json
cayu guide applications#planning --json

Cloud commands ship in the same cayu package; no additional CLI package is required. Cayu Cloud is currently invite-only. Login-backed commands are pinned to the production service at https://cloud.cayu.dev; users never select an API URL.

cayu cloud --help
cayu cloud login
cayu cloud whoami
cayu cloud init
cayu cloud deploy

Login uses WorkOS device authorization. It opens the browser when possible and always prints a verification URL and one-time user code for SSH, containers, Cursor, Codex, and Claude Code. The resulting Organization-scoped session is stored privately and refreshed automatically; no WorkOS secret is embedded in Cayu. Use --no-browser when a human will open the displayed URL on another device. Login selects the WorkOS session over any previously persisted private context; a later cayu cloud context use PATH deliberately switches back to that internal automation context. Private contexts and explicit API credentials remain endpoint-bound operator mechanisms and do not change the default customer endpoint.

cayu cloud init generates the small deployment descriptor from standard pyproject.toml metadata and a configured Cayu server, worker, or console script. Review the generated process topology before deploying it. Existing descriptors are never replaced unless --force is explicit.

The default command packages the current local working directory and uploads it directly to the selected Cayu Cloud Organization as an immutable source bundle. It includes an applied patch and does not require GitHub, a clean worktree, a commit, or a push. Git-ignored files and common local credential/cache paths are omitted.

Deploy creates or updates the 8-63 character application slug declared in cayu-cloud.toml. Slugs use lowercase letters, numbers, and interior hyphens. --application SLUG overrides that create-or-update slug; check it carefully because a valid typo creates a separate application.

Deploy output and local evidence redact runtime environment values. Cayu verifies the evidence destination before Cloud mutation; if only the final evidence write fails after a successful rollout, deploy still exits successfully with evidence_id: null and an evidence.status: unavailable diagnostic.

To deploy an exact remote GitHub revision instead, pass its canonical URL and 40-character commit. For a private source, make authentication available to the process that runs Cayu:

gh auth status --hostname github.com
# For noninteractive CI or an isolated Codex/Claude Code home:
GH_TOKEN="$GITHUB_TOKEN" cayu cloud deploy \
  https://github.com/example/agent --revision COMMIT_SHA

GitHub CLI credentials stored in a desktop keychain may not be reachable after changing HOME, even when GH_CONFIG_DIR points at an authenticated GitHub CLI configuration. In that case, pass a short-lived GH_TOKEN explicitly. Cayu does not persist the token in its Cloud context or include GitHub CLI stderr in its JSON errors. Public GitHub repositories remain deployable without GitHub authentication.

For CI and other noninteractive automation, the existing CAYU_CLOUD_API_KEY, CAYU_CLOUD_API_KEY_FILE, and private Cloud-context options remain available. Explicit automation credentials take precedence over the saved interactive login.

This compact example shows the core API. Real projects should put the same registrations in the generated build_app() factory instead of constructing a module-global app.

import asyncio

from cayu import AgentSpec, CayuApp, Message, RunRequest, run_to_completion
from cayu.providers import OpenAIProvider


async def main() -> None:
    app = CayuApp()
    app.register_provider(OpenAIProvider(), default=True)  # reads OPENAI_API_KEY
    app.register_agent(AgentSpec(name="assistant", model="gpt-5.6"))

    outcome = await run_to_completion(
        app,
        RunRequest(
            agent_name="assistant",
            messages=[Message.text("user", "Explain durable agent sessions.")],
        ),
    )

    if outcome.ok:
        print(outcome.final_text)
    else:
        print(f"{outcome.status}: {outcome.error}")


asyncio.run(main())

CayuApp() uses in-memory stores by default, which is appropriate for this one-shot example and for tests. The generated project configures all local Cayu stores in data/cayu.db so sessions survive process restarts. Multi-process production deployments should select a conforming shared store such as PostgreSQL.

CayuApp.run(...) is the lower-level event-stream API. Runtime failures arrive as terminal session.failed events instead of exceptions raised from iteration. run_to_completion(...) consumes that same stream and returns a typed outcome when an application only needs the result. It retains the complete event stream in RunOutcome.events; use it for bounded runs. Consume CayuApp.run(...) incrementally for long-lived or high-volume runs.

For a credential-free domain-tool tracer bullet, run cayu guide references#domain-tool, then use cayu generate tool. To add workspace tools and command execution, see examples/local_environment_runtime.py. For a finite application-owned command surface, use NamedCheck and RunCheckTool instead of exposing model-authored shell or argv.

The generated AGENTS.md is the project-local source of truth. Ask the coding agent to read it first, then use Cayu's package-shipped guides and structured inspection:

cayu guide anatomy
cayu guide authoring
cayu inspect --json
cayu check --fail-on warning --json

The package-shipped cayu guide authoring#cayu-map routes each optional capability to the smallest version-matched local reference. Its online source mirror is secondary. The examples index provides runnable references without making them required project structure.

The supported authoring loop is:

understand -> inspect -> plan -> change -> test -> eval -> exercise -> report evidence

Start by editing the existing model-only agent, test, and eval. Add a generated tool-backed slice only when the requested job needs a capability outside the model; generated slices remain unfinished until their placeholder behavior, test, and eval have been replaced.

Cayu separates the agent's identity from the resources and durable state used for one execution:

AgentSpec
  identity, model, system prompt, defaults, runtime policies

Environment
  workspace, runner, artifacts, vault, proxy, knowledge, MCP

Session
  durable identity, transcript, events, status, checkpoints

ToolContext
  the active environment services and call identity for one tool execution

• Agent describes who is acting and how model work is configured.
• Environment describes what that agent can touch.
• Session records one durable execution and its lineage.
• Tool is an explicitly registered, application-owned capability that the model may request. A native Python Tool runs inside the trusted Cayu application process; ToolPolicy gates its invocation but does not sandbox its implementation.
• Task is an optional durable unit of background or orchestrated work.
• Workflow is deterministic application orchestration around agent steps.

An environment is optional for a conversational agent. It becomes important when tools need files, commands, artifacts, secrets, network policy, or a sandbox. Static environments are useful for trusted local work; EnvironmentFactory creates or reattaches session-specific environments in production.

Choose the execution surface according to where code should run and which boundary should contain it:

Do not add every Cayu primitive to every application.

Start a conversation agent with the model and state it needs. Add workflows, task queues, environments, memory stores, servers, or multi-agent topology when the behavior requires them. Give coding agents narrow domain tools before granting broader shell access.

Artifact writes have a bounded cancellation-settlement contract. After a dispatched write is interrupted, stores distinguish an exact committed artifact, positively proved absence, and a typed reconciliation-required orphan candidate; the candidate is diagnostic evidence, never a readable artifact reference.

Your application should own:

• end-user prompts and domain forms;
• product authentication and authorization;
• business-specific workflow and state;
• user-facing streaming, notifications, and presentation; and
• decisions about when a run, task, approval, or interruption is allowed.

Cayu owns runtime execution and the operational state recorded by the application's configured stores. Its optional dashboard is a control plane for developers and operators: inspect sessions, events, transcripts, tasks, usage, artifacts, pending actions, and recovery state. Your application remains responsible for the product experience.

Start work through the API that matches the trigger:

• run for an immediate new session;
• resume for a deliberate continuation;
• dispatch for placement through a dispatcher;
• a task worker for durable queued work;
• a subagent for model-selected bounded delegation; or
• an event watcher for durable reactions to already-persisted events.

See Triggering runs for the decision guide and lifecycle responsibilities.

The base package includes the provider contracts and built-in OpenAI, Anthropic, OpenAI-compatible HTTP, and experimental OpenAI-subscription adapters. Optional extras add integrations without forcing their dependencies into every deployment:

cayu[all] intentionally excludes cayu[browser], which is dashboard and browser verification tooling rather than a runtime integration.

Providers normalize text, thinking, tool calls, usage, completion reasons, and typed failures behind one runtime contract. Applications register providers explicitly and may add deterministic model-pattern routing; an arbitrary model name never selects a provider.

Cayu focuses on OpenAI, Anthropic, OpenRouter, Google, Bedrock, and Vertex. OpenRouter is a first-class cayu new --provider openrouter and CAYU_PROVIDER=openrouter choice backed by the generic Chat Completions adapter; it requires OPENROUTER_API_KEY and an explicit CAYU_MODEL slug. Compatible Chat Completions services such as Fireworks, Baseten Model APIs, and OpenCode Go use that generic adapter directly. Run cayu guide providers#openrouter or use the package guide for exact setup.

For native Choice, Score, and Noul decisions, explicitly register the experimental TypeSafeProvider. It uses TypeSafe's System One API through Cayu sessions and events; it does not support chat tools or multimodal inputs.

For local development without separate OpenAI API billing, users can sign in with their own ChatGPT subscription:

cayu auth openai login
# For SSH or a remote machine:
cayu auth openai login --headless

from cayu.providers import OpenAISubscriptionProvider

app.register_provider(OpenAISubscriptionProvider(), default=True)

This experimental integration uses the Codex backend. It does not use the documented OpenAI Platform API. Cayu identifies itself with originator: cayu and preserves upstream rejections. OpenAI has not documented this raw backend as a general third-party provider API, so support may change or stop.

Intended-use boundary: Use this path only for a subscription holder's own local development and evaluation. For production, customer-facing or multi-user services, use the OpenAI Platform API or another officially supported provider. Do not share or resell credentials or bypass plan limits.

See OpenAI subscription authentication for the support boundary, credential storage, and fallback options.

The same agent can run in a local workspace, trusted Docker container, E2B, Microsandbox, Lambda MicroVM, or an application-owned runner without changing its identity or transcript contract.

Cayu ships three equally supported control-plane choices: use the bundled compiled dashboard, eject the exact version-matched React/TypeScript source for application-owned customization, or provide a completely custom UI over the versioned control-plane API.

cayu dashboard eject ./control-plane
cd control-plane
npm ci
npm run dev
npm run build

Extraction uses package data only—no repository clone, GitHub access, or network request. Serve the resulting dist/ with DashboardConfig(directory=...), mount_cayu(..., dashboard_dir=...), or mount_dashboard(..., dashboard_dir=...). Cayu never rewrites the extracted application-owned source during upgrades. See Open and replaceable control plane for the complete workflow, compatibility gate, and redistribution obligations.

Cayu makes safety boundaries explicit, but configuration still matters:

• Native Python Tool implementations are trusted host-process code and can access authority available to the Cayu application. ToolPolicy controls whether the model may call a tool and with which arguments; it is not an OS isolation boundary. Run model-authored or otherwise untrusted code through an admitted runner or separately governed external tool boundary. Native tools that need credentials should use explicit SecretRef values through ctx.proxy or ctx.vault and return only safe results; ambient host environment values are not automatically mediated as workload credentials.
• LocalRunner executes directly on a trusted local machine and provides no sandbox isolation.
• DockerRunner is useful for development and CI; ordinary Docker isolation is not presented as a secure untrusted-code boundary.
• Environment registration does not imply selection: mark a default explicitly or name the environment on the request. Provider defaults and model-pattern routing should likewise be configured deliberately and kept unambiguous.
• Tool effects do not authorize themselves. Use policies, approvals, scoped credentials, and destination controls where consequences require them.
• SQLite is appropriate for local and single-writer deployments. Use PostgreSQL or another conforming shared store for sustained multi-process concurrency.
• The FastAPI control plane requires an explicit ServerConfig access policy. Use AuthenticatedAccess for deployed operator surfaces; OpenAccess and ServerConfig.local_development() are deliberate local-only choices. Deployment names are descriptive metadata and never relax security policy. See server configuration. AuthContext.tenant records authenticated operator provenance but does not filter or isolate Cayu data. See Server authentication and tenant isolation. Generated API documentation is a separate exposure decision.
• Public or multi-user product routes need a separate customer authorization boundary; operator authentication does not make the raw Cayu surface tenant-scoped. Start new services with cayu new NAME --preset service, then require both cayu check --deploy --fail-on warning --json and the generated tests/test_public_service_security.py. Cayu reports arbitrary host-owned ASGI routes outside that maintained factory as unverified.
• When embedding with mount_cayu(..., path="/your/path") or the lower-level mount_dashboard(...), use /your/path/ as the canonical dashboard URL. Cayu redirects an exact GET or HEAD of the slashless non-root mount after a successful dashboard mount. That public 307 may be returned without credentials; dashboard HTML, assets, deep links, and other protected content at the canonical target still require configured authentication. mount_cayu(...) places its control-plane API under /your/path/api; mount_dashboard(...) configures apiBaseUrl independently and defaults it to /api.
• Usage is derived from recorded events and survives restarts when those events use a durable store; cost remains an estimate against the price book your application selects.
• Recovery never invents the outcome of an ambiguous external side effect. Reconcile it through the typed recovery APIs.

Read Runtime contracts before changing persistence, replay, approval, interruption, budget, provider, runner, or recovery behavior.

Start with the documentation index for maintained guides, operational references, design records, research evidence, and explicitly archived material.

Start with the document that matches the job:

Maintainer-facing architecture is documented in Architecture, Project layout, and the Glossary.

• Examples index: find the smallest reference for a capability.
• cayu guide references#domain-tool: credential-free domain-tool authoring and generator path.
• Local environment runtime: files and commands.
• Server example: protected API and control plane.
• Cloud PR reviewer: durable task, isolated workspace, QA, and an explicit external effect.
• Business approvals: domain approval routing over the binary runtime primitive.
• GitHub CLI through virtual egress: an unmodified CLI with a virtual token, exact REST policy, and explicit mutation boundary.
• Advanced runtime examples: forks, compaction, taint isolation, speculative approval, and measured evidence.

Advanced examples are executable runtime specifications. Each example states its evidence boundary instead of presenting one strategy as suitable for every workload. Their measured results are described in Advanced runtime strategies.

Cayu contributors should read CONTRIBUTING.md for placement policy, setup, validation commands, and pull-request requirements. New third-party integrations normally live in their own packages against Cayu's public extension contracts.

Report suspected vulnerabilities privately as described in SECURITY.md. Do not open a public issue or pull request for a suspected security vulnerability.

For questions and project discussion, join Discord. Use GitHub issues for actionable bugs and concrete feature proposals.

Cayu is licensed under the Apache License 2.0.