Okou
Delegate cross-app work in natural language while keeping permissions, approvals, and execution under team control.
- Source repo
- okou-ai/okou
- Stars
- ★ 1.2k
- Last updated
- today
- License
- NOASSERTION
- Primary language
- TypeScript
- FA score
- 54/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformOpenAI API · Claude API
- Cost
- Free tier plus a paid hosted plan
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A founder starts the day with a brief assembled from Calendar, Linear, Slack, and X, then has investor updates drafted from live metrics.
- Not a fit if
- Teams that need fully offline or on-premise operation
- Users unwilling to run workflows and credentials in the vendor cloud
- Teams that need clear license terms before self-hosting
- Source review
- 54/100 · Major gaps
What does this agent do, and when should you use it?
Okou is a workflow agent for marketing, sales, engineering, and operations teams, accepting job-level requests through chat or Slack instead of requiring step-by-step prompts. It determines the steps, selects connected services, and can run several independent task chats concurrently. A completed run can be saved with its instructions and tools as a workspace-owned workflow, then launched by another member, a schedule, or a trigger. Each member runs shared workflows with their own service permissions, while actions that publish externally or spend money stop as drafts or approval requests. Tasks execute in isolated Firecracker microVMs with logs, metrics, and network visibility; the project also claims self-hosting and multi-provider model selection, although the supplied material does not document deployment commands or infrastructure requirements.
A user assigns a job in chat or Slack, after which Okou works out the steps and calls authorized connected services. Named integrations include GitHub, Gmail, Google Calendar, Google Sheets, Notion, Linear, Sentry, HubSpot, Intercom, Figma, Vercel, Dropbox, Airtable, Meta Ads, X, and Reddit, alongside a claimed catalog of more than 3,000 connectors. Documented examples include reading brand material or operational data, publishing a web page, updating CRM records, grouping Sentry events by root cause, filing issues, and producing reports, decks, release notes, and email drafts. Each job receives its own chat and isolated Firecracker microVM, allowing multiple jobs to run at once. Repeated runs can become workspace workflows invoked from chat or Slack and started by schedules or triggers; outbound and spending actions remain subject to human approval.
- A founder starts the day with a brief assembled from Calendar, Linear, Slack, and X, then has investor updates drafted from live metrics.
- A sales or marketing team researches prospects and creators, writes the findings to a tracker, and drafts personalized outreach based on each prospect's content.
- An engineering team groups Sentry noise by root cause, files scoped issues, and generates release notes from merged pull requests.
- An operations lead produces weekly status reports, onboarding checklists, and meeting digests with owners, then schedules follow-up on approvals and deadlines.
- A cross-functional launch team runs landing-page publication, CRM updates, a newsletter draft, and a metrics deck as separate concurrent tasks.
- A teammate who has validated a recurring process saves it as a workspace workflow so colleagues can run it with their own permissions or automate it with a trigger.
How do you install or deploy this agent?
The supplied material provides no local installation commands, self-hosting procedure, dependency versions, configuration files, environment variables, or startup command, so a reproducible self-hosted installation cannot be given. It only documents the hosted signup page at https://app.okou.ai/sign-up and a Slack installation endpoint at https://www.okou.ai/api/slack/oauth/install?publicBrand=okou. Connected services require appropriate accounts and explicit authorization, but credential formats, OAuth scopes, and Firecracker host requirements are not specified.
How do you use this agent?
Sign up for the hosted service or add Okou to Slack, connect the services required for the job, and grant suitable read or write permission for each tool. Assign the desired outcome conversationally—for example, ask it to build a one-page site using a brand kit in Drive—without prescribing every intermediate step. Review the separate task chats, tool activity, and produced artifacts; approve controlled actions such as sending email or changing an advertising budget only after inspection. For recurring jobs, save the run's instructions and selected tools as a workspace workflow, then let teammates invoke it or attach a schedule or trigger. The material does not document the connector configuration screens, retry controls, or a first invocation for self-hosted deployments.
What are this agent's strengths and limitations?
- Independent task chats can run concurrently, supporting launches that combine web publishing, CRM updates, email drafting, and presentation work.
- Saved workflows preserve both instructions and selected tools as workspace assets, turning an individually discovered process into a reusable team operation.
- A shared workflow executes with each member's own permissions instead of borrowing access from its creator.
- Outbound communication and spending changes stop at human approval points, complemented by per-tool and per-agent read/write controls.
- Every task runs in an isolated Firecracker microVM, with logs, metrics, and network visibility for auditing.
- Anthropic, OpenAI, DeepSeek, Google, and other model providers can be selected per run or per agent, reducing dependence on one model vendor.
- No installation commands, version requirements, architecture guide, or self-hosting runbook are present in the supplied material, making deployment effort impossible to verify.
- The connector list does not define the exact operations, permission scopes, or feature coverage available for each service.
- Firecracker microVMs and network access are central to the execution model, but host compatibility, capacity planning, and operating-system requirements are undocumented.
- Pricing, hosted usage limits, model costs, and possible fees for the claimed 3,000-plus connectors are not disclosed.
- Retry behavior, rollback, checkpoint recovery, and handling of partial connector failures are not documented.
- The practical terms for open-source use and self-hosting cannot be established from the repository metadata because the license is marked NOASSERTION and the referenced LICENSE text was not supplied.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Okou This agent | 54 · Major gaps | Hosted serviceFreemium | ★ 1.2k | today | TypeScript | OpenAI API · Claude API |
| iloveAgents | 41 · Major gaps | Web appFree + model costs | ★ 150 | 8d ago | JavaScript | OpenAI API · Claude API |
| Skales | 55 · Major gaps | Desktop appFreemium | ★ 2.2k | 6d ago | — | ChatGPT · OpenAI API · Claude API |
| Atom Platform | 67 · Some gaps | CLIFree + model costs | ★ 911 | today | Python | OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
The README says permissions are configured per tool and agent, tasks run in isolated sandboxes, and runs expose logs and network visibility; its example also leaves email unsent. These support least privilege, confirmation, and external-effect controls, though mostly as product assertions. Sensitive-data handling has some concrete support in the Guest Agent process-inspection restriction and Runner credential-boundary documentation, but the sample does not establish complete data flows or credential lifecycles. Dependency security is thin: one pinned boto3 version and pinned Action SHAs appear, but there is no repository-wide dependency-security evidence. Rollback appears only in a partial engineering-document excerpt, with no clear product-user recovery path. Attribution is limited to the licensor named in LICENSE; publisher identity remains unverified.
The project guidance, Runner configuration excerpt, and Guest Agent error-handling excerpt show consistent compatibility, cleanup, and failure-handling principles; the README also describes parallel tasks and auditable runs. Dependency availability is weakly evidenced: many connectors and model providers are listed, but the sample gives no availability commitments or outage behavior. Failure handling has concrete examples and fail-closed engineering guidance, but not enough evidence to establish end-to-end behavior across integrations.
The README names founders, sales and marketing, engineering, and operations/support, with scenarios for each; per-user permissions and per-agent tool permissions show some capability boundaries. It mentions chat, Slack, schedules, and triggers, but gives little precision about trigger conditions or automation boundaries. Environment-fit evidence is limited: the Runner excerpt covers engineering configuration, while setup guidance for ordinary users is sparse.
The README has clear product sections, scenario examples, and a license pointer; sampled skills use names, descriptions, and task routing, and engineering guidance has an index. Installation notes are limited, FAQ and known limitations are incomplete, and SECURITY.md says only the latest release is supported. LICENSE clearly states BSL 1.1, its additional-use restriction, a four-year change date, and Apache 2.0 as the change license, so license disclosure is adequate, although the commercial Agent Infrastructure Service restriction matters. The sample provides no clear changelog or update path. SECURITY.md specifies private reporting, response targets, and disclosure steps, supporting some maintenance-responsibility credit; publisher identity is still unverified.
The README shows usable outputs ranging from drafts and published work to shared workflows, across several business functions and tools. That indicates potential practical value. However, examples and connector counts are product claims, and the sample does not establish output quality, cost, or operating limits, so marginal value and cost-benefit scores remain limited.
Specific README examples and claims about permissions and isolation can be partially compared with the Guest Agent and Runner material. Many product capabilities, security controls, and connector-coverage claims remain README-only, without implementation evidence or independent corroboration. The documents distinguish some behavior, configuration conditions, and limitations, including Runner memory overhead and budget risks, so fact/inference separation is reasonably handled; a complete traceability chain for product-level claims is absent.
- Most evidence for security, isolation, auditability, connector coverage, and automation comes from README claims; this static review does not confirm runtime behavior.
- The BSL license restricts use of the work for an Agent Infrastructure Service serving third parties; check the license scope before deployment or redistribution.
- The sample does not provide complete installation, cost, integration-failure, or rollback guidance; files not shown are unknown.
Maintain this project? Add the FARS badge to your README
[](https://followagents.com/en/agents/okou)<a href="https://followagents.com/en/agents/okou"><img src="https://followagents.com/badge/okou.svg" alt="FARS review"></a>Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
runner-rs-v0.244.44 runner-rs: v0.244.44
Release notes excerpt
[0.244.44](https://github.com/okou-ai/okou/compare/runner-rs-v0.244.43...runner-rs-v0.244.44) (2026-11-08) Release Dependencies * Release of turbo/apps/cliView on GitHub ↗ -
pi-agent-runtime-v1.51.86 pi-agent-runtime: v1.51.86
Release notes excerpt
[1.51.86](https://github.com/okou-ai/okou/compare/pi-agent-runtime-v1.51.85...pi-agent-runtime-v1.51.86) (2026-11-08) Dependencies * The following workspace dependencies were updated * devDependencies * @okouai/core bumped to 8.824.2View on GitHub ↗ -
runner-rs-v0.244.43 runner-rs: v0.244.43
Release notes excerpt
[0.244.43](https://github.com/okou-ai/okou/compare/runner-rs-v0.244.42...runner-rs-v0.244.43) (2026-11-08) Refactoring * python: share iterable wrapper recognition in metadata linting ([#45959](https://github.com/okou-ai/okou/issues/45959)) ([570245d](https://github.com/okou-ai/okou/commit/570245d6e74d5d7fad03ce460758481a48eaaeaf)) Release Dependencies * Release of turbo/apps/cliView on GitHub ↗
FAQ
Will Okou send emails or change advertising budgets without approval?
Does a shared workflow reuse its creator's account permissions?
Can a team choose its model provider?
Can Okou be self-hosted?
What happens when a task or connector fails?
Thanks — we use this to improve the page.