Okto Nexus Coordination Hub

Coordinate AI agents working in one repository with local-first messaging, durable handoffs, and operator oversight.

Stars
★ 110
Last updated
10d ago
License
NOASSERTION
Primary language
Python

At a glance

Works with
Universal · cross-platformCodex · Claude Code (Partial support)
You'll need
Python >=3.11SQLiteMCP clientShell / CLINetwork accessLocal filesystemMCP Server
Typical use
A software team coordinating agents in one repository can use sessions, presence, and inboxes to keep work discoverable.
Main limitation
The project calls itself a development release and says final G0–G3 acceptance is still open.
Website
oktolabs.ai/platform/nexus ↗ set on GitHub
Maintenance
Commits in the last 30 dayslast commit 2026-11-10latest release v0.1.3 (2026-07-16)
Source review
77/100 · Good

What does this agent do, and when should you use it?

Okto Nexus is an MCP server and operator hub for agent teams working in the same repository, with identities, presence, inboxes, handoffs, artifacts, an event log, governance controls, and a live dashboard. okto-nexus serve exposes a FastAPI-based HTTP service on one port: /mcp uses streamable HTTP, /api/v1 provides operator REST APIs, read-only monitor endpoints, and SSE, and / serves the bundled React dashboard. Coordination records live in a local SQLite database in WAL mode, while artifact payloads use an adapter-backed store; no cloud broker is required. It requires Python >=3.11 and can be installed as a package or run from a checkout. The repository describes this as a development release and says final G0–G3 acceptance remains open.

An operator starts okto-nexus serve, creates agent identities in the dashboard, and copies each nxs_... API key when it is shown once. An MCP client sends that key in the HTTP Authorization: Bearer header to /mcp, then calls agent_whoami(), workspace_resolve(project_root=...), and session_open(...) before checking its inbox and anchoring an event cursor. Agents use message_create for messages delivered to durable recipient inboxes and handoff_create for work that one eligible agent can claim; recipients pull and acknowledge messages, while an executor claims and completes a handoff. The service records coordination state in SQLite and emits events; operators inspect activity through /api/v1, SSE, okto-nexus tail, and the dashboard. Artifact payloads are stored by the local file adapter.

  1. A software team coordinating agents in one repository can use sessions, presence, and inboxes to keep work discoverable.
  2. An engineering lead who needs visible ownership can assign executable work as handoffs with leases, claims, and completion states.
  3. A team running agents on a local machine or controlled network can keep coordination records in SQLite and inspect events in the dashboard.
  4. An operator who needs to restrict agent writes can apply permissions, versioned policies, guardrails, and optional human approval.
  5. A team giving a helper process limited monitoring access can issue short-lived nxsept_... tokens scoped to documented event and inbox endpoints.

How do you install or deploy this agent?

The recommended install includes the HTTP hub, dashboard, local embedding provider, and tokenizer. Python >=3.11 is required.

How do you use this agent?

After starting the server, create an identity under Agents → New agent and save its one-time key. Configure an MCP client to use the local HTTP endpoint, then resolve the workspace, open a session, and check the inbox on first use.

What are this agent's strengths and limitations?

Pros
  • A single okto-nexus serve process provides MCP HTTP, REST, SSE, and the React dashboard.
  • Messages have durable per-recipient inboxes, leases, redelivery, and acknowledgements; handoffs support atomic claims.
  • Governance includes permissions, communication scopes, versioned policies, guardrails, and optional human approval.
  • Coordination records use SQLite without a cloud broker, and artifact payloads use a separate adapter-backed store.
Limitations
  • The project calls itself a development release and says final G0–G3 acceptance is still open.
  • It is designed for local or controlled single-tenant coordination, not a public multi-tenant broker; SQLite is the only built-in coordination store.
  • The server does not terminate TLS; remote dashboard access requires an authenticated TLS reverse proxy in front.
  • Session secrets are stored in plaintext in the local SQLite database, putting anyone who can read that file inside the session trust boundary.

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
Okto Nexus Coordination Hub This agent 77 · Good — ★ 110 10d ago Python —
Guild 62 · Some gaps CLIFree ★ 297 11d ago Go Codex · Claude Code
Mnemo Cortex 75 · Good Self-hosted serviceFree ★ 161 1d ago Python Codex · Claude Code · Claude.ai
Agents Observe 48 · Major gaps Agent plugin / skillFree ★ 738 2mo ago TypeScript Claude Code

How does FollowAgents rate this agent?

FollowAgents source review · FARS-2.1
Good
77/ 100 5-point scale 3.9 / 5
Trust 19/29
Reliability 8/14
Adaptability 16/18
Convention 15/18
Effectiveness 12/13
Verifiability 7/8
Why each dimension lost points
Trust19 / 29 · 3.3/5

The README describes separate operator and agent authentication, scoped short-lived monitor tokens, hashed API keys, permission and approval interception, and where SQLite, external files, and derived views are stored. This supports strong scores for least privilege, data-flow transparency, and control of external effects. Human approval exists for an interception flow, but ordinary write confirmation and recovery coverage are not fully shown. Key hashing and report-redaction guidance support sensitive-data handling, though they do not establish comprehensive protection. Dependencies have version ranges and the connector core is pinned exactly, but the packet does not show dependency vulnerability handling. The license requires preserving branding; publisher identity is unverified and source verification is not clearly documented, so attribution is only partial.

Reliability8 / 14 · 2.9/5

The documentation describes SQLite WAL, migrations, leases, redelivery, recovery, and error codes; test samples also show an isolated real-server fixture and crash scenarios. However, the README says version 0.3.0 while pyproject.toml says 0.3.2. CLAUDE.md says there is no CI and lists a retired stdio entry point, conflicting with the README and supplied CI workflow. These conflicts lower self-consistency. Installation options, optional features, and behavior when optional dependencies are absent are explained. Error codes and stages are described, but this sample does not include a complete error reference.

Adaptability16 / 18 · 4.4/5

The README clearly targets multi-agent teams coordinating in a shared repository and describes messaging, handoffs, broadcast, monitoring, and local or controlled single-tenant use. It distinguishes identity, capability-based routing, and authorization, and documents MCP, REST, SSE, dashboard, and optional embedding boundaries. Guidance for choosing a handoff versus an informational message is precise. Python requirements, install options, host binding, and remote access are documented, though the default binds all IPv4 interfaces and deployment security remains an operator responsibility.

Convention15 / 18 · 4.2/5

The README has a thorough contents map and sections for architecture, configuration, operations, errors, troubleshooting, and security. Install commands, MCP setup, workflow examples, and limitations are detailed. The license text and package metadata explain restrictive use and branding terms. Version evidence conflicts, and the sample lacks a complete release history, so version stability is only partial. SECURITY.md provides a vulnerability-report address and assigns upgrade responsibility but promises no response time; with publisher identity unverified, maintenance ownership evidence is limited.

Effectiveness12 / 13 · 4.6/5

Durable inboxes with leases and acknowledgements, single-winner handoffs, event cursors, SSE, and an operator dashboard provide concrete coordination functions with documented outputs and consumption paths. The distinction between messages and work-producing tasks makes the intended use clearer. The complete service includes a heavyweight local embedding model and tokenizer, while a lighter install is offered; resource costs and measured benefits are not quantified, so cost-benefit does not merit full marks.

Verifiability7 / 8 · 4.4/5

The README makes specific, checkable claims about interfaces, defaults, authorization boundaries, leases, and retention. The package metadata, CI configuration, and test fixtures corroborate some installation and engineering claims. Conflicting version and CI/transport statements weaken cross-source corroboration. The documentation generally separates implementation descriptions, development status, and outstanding qualification, including explicit statements that some integration gates remain open. This assessment uses only the supplied static material and does not treat claims as executed verification.

Risks and how to mitigate them
  • The README and pyproject.toml give different version numbers; CLAUDE.md statements about CI and stdio transport conflict with the supplied README and CI workflow.
  • The default bind address is 0.0.0.0. Check access controls and TLS before remote deployment. SECURITY.md assigns host, filesystem, secret, and timely-upgrade responsibilities to the operator.
  • The license restricts hosted and competing services and requires visible branding; review its full terms before adoption.
Evidence confidence: Low Reviewed Nov 11, 2026 Reviewed revision 8d77b50c00b4
See the full review method →
Maintain this project? Add the FARS badge to your README

The badge shows the current review score and verdict and links back to this page. It refreshes about once an hour after a re-review.

FARS

[![FARS review](https://followagents.com/badge/okto-nexus.svg)](https://followagents.com/en/agents/okto-nexus)
<a href="https://followagents.com/en/agents/okto-nexus"><img src="https://followagents.com/badge/okto-nexus.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.

  1. v0.1.3 Okto Nexus 0.1.3
    Release notes excerpt
    Maintenance, documentation, and repository-hygiene release. The MCP surface remains at revision 31 and the database schema remains at migration 026. Highlights: - Completely rewritten README aligned with the current transports, authentication, dashboard, 43/46-tool MCP surface, routing, retention, governance, workflows, schema, and operations. - Package and lock metadata synchronized at 0.1.3. - Dashboard/package license label aligned with the bundled SaaS/Branding Addendum. - Runtime SQLite databases and sidecars remain excluded from release archives. …
    View on GitHub ↗

FAQ

Does it require a cloud message broker?
No. Coordination records are stored in a local SQLite/WAL database; the project documents no Redis, PostgreSQL, or cloud broker adapter.
How does an MCP client authenticate?
It sends an agent API key in the HTTP Authorization: Bearer nxs_... header. New configurations should keep the key out of the URL.
Can I expose the service directly to the public internet?
The project is intended for local or controlled single-tenant use, not as a public multi-tenant service. It does not terminate TLS, so remote access should use an authenticated TLS reverse proxy.
Are messages and artifacts retained indefinitely?
No. Messages are removed according to configured retention; artifact payloads live in managed local storage, and explicit database resets also affect stored data.
Which optional features are on by default?
All seven optional feature flags default to off, including memory, approval interception, verification, DAG dependencies, health, trace, and REST replay. Memory tools require enabling the flag and restarting.
Is Okto Nexus Coordination Hub open source?
The source code is public on GitHub (OktoLabsAI/okto-nexus), but the repository does not declare a recognised licence, so the rights to reuse or modify it are unclear. Ask the author or check the README before building on it.
Is Okto Nexus Coordination Hub safe to use?
FollowAgents' FARS review (Nov 11, 2026) scored Okto Nexus Coordination Hub 19/29 on trust and safety. None of the eight safety controls the review checks for was missing outright.
Is Okto Nexus Coordination Hub still maintained?
Yes. The latest commit on GitHub is from 2026-11-10, within the last 30 days. The latest release is v0.1.3 (2026-07-16).
Who makes Okto Nexus Coordination Hub?
Okto Nexus Coordination Hub is published on GitHub by the account OktoLabsAI. FollowAgents has not verified which person or company is behind that account.
What is Okto Nexus Coordination Hub's official website?
The project's GitHub repository lists https://oktolabs.ai/platform/nexus/ as its website. That link is set by the repository owner; the source code itself is on GitHub (OktoLabsAI/okto-nexus).
Was this page helpful?
View on GitHub ↗ Install ↓

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related concepts