Hyperspell vs SageOx (2026)

More than 50 OAuth connectors, from Slack, Gmail and Notion to Salesforce, HubSpot, Jira and GitHub, bring content into the brain. Permissions are inherited, and each connection has its own channel allowlists and folder policies. A live mode reads a source at query time, and each connector searches or queries its source. Hyperspell names no connector that writes back so an agent can act in the outside system.

SourceHyperspell, docs.hyperspell.com integrations/overview and hyperspell.comread 2026-09-21

The Slack integration, authorized for each team, publishes into the channels a team picks (discussion murals, live join links, team murals and welcome notes). SageOx listens in Zoom and Google Meet calls it is invited to and connects to GitHub or GitLab with the access granted at setup, and its open-source AI coworker answers in the Slack thread that asked and can wake a named person. These are native, authorized write actions in real systems. Slack listening is marked not yet available.

SourceSageOx, sageox.ai/docs integrations, integrations/slack and your-data, and sageox.ai changelogread 2026-09-21

Agents query Hyperspell as a context layer. It does not sequence, branch or route their workflows.

SourceHyperspell, docs.hyperspell.com agents/overviewread 2026-09-21

SageOx works inside agents that other tools launch and orchestrate (OpenClaw, Factory Droid or custom orchestration), and its own surfaces prime, capture and plan. It runs no workflows that sequence, branch or retry steps, or that combine deterministic nodes with agent steps.

SourceSageOx, sageox.ai/docs agent-factories, how-it-works and plansread 2026-09-21

Agents query Hyperspell when they need context. Continuous indexing and real-time message ingestion keep the brain fresh. Hyperspell names no schedule, event or channel that starts an agent's work on its own.

SourceHyperspell, docs.hyperspell.com agents/overview and api-reference/messages/ingest-messagesread 2026-09-21

Scheduled jobs run SageOx's AI coworker, including jobs whose body is a prompt so the coworker does light scheduled work on its own, and it answers mentions and missed DMs in Slack. In repositories, agent lifecycle hooks start session capture automatically, and there is a scheduled workflow for daily distillation and queries in CI.

SourceSageOx, sageox.ai changelog and sageox.ai/docs cli/ephemeral-mode and claude-coderead 2026-09-21

More than 50 sources such as Slack, Gmail, Notion, Linear, HubSpot, Salesforce and Google Drive connect over OAuth and are continuously indexed into one company brain. Content is stored once with a per-document list of who may see it. Agents query the brain for grounded answers with a trail back to the source, and a live search reaches content not yet indexed.

SourceHyperspell, docs.hyperspell.com core/permissions, agents/overview and hyperspell.comread 2026-09-21

Team documents added to Team Context are indexed automatically, so repo-based AI coworkers see a catalog at session start and read full content on demand, and MCP clients search and retrieve the same knowledge. The ox query command runs semantic search across Team Context, CodeDB builds a local semantic index of the codebase and its git history for coworkers to query, and plan badges cite the real artifact behind each claim. That is a structure over the customer's knowledge that is kept current and cites its sources.

SourceSageOx, sageox.ai/docs team-context, codedb, cli/query and plansread 2026-09-21

Agents can write their traces and transcripts, real-time messages and manually added memories into the index. These persist across sessions under the same per-person permissions, and memories or a user's whole data set can be deleted. Hyperspell states no lifetime for stored memories and names no memory scoped to an agent or workflow, separate from the company index.

SourceHyperspell, docs.hyperspell.com api-reference/traces/add-an-agent-trace, integrations/all/vault and api-reference/authentication/delete-userread 2026-09-21

Team Context keeps a memory directory of team memory organized in time-based layers, built automatically from recorded discussions and observations, with an auto-generated MEMORY.md index that repo-based AI coworkers load at session start. It is team-wide, held in a git repository the team can browse, edit, commit to and diff, and built-in redaction runs before observations are stored, with team-set redaction rules. So the memory has a clear scope and a time based lifetime, and the team can review, edit and delete it.

SourceSageOx, sageox.ai/docs team-context and your-dataread 2026-09-21

Context flows from Hyperspell to agents that act elsewhere. Hyperspell names no approval step, escalation rule or pause for a person's sign-off before an agent takes an action.

SourceHyperspell, docs.hyperspell.com api-reference/memories/approve-a-pending-resource and advanced/content-rulesread 2026-09-21

An AI coworker writes a visual plan that a person reviews and decides go or no-go on, and the coworker revises the same page on feedback before the work becomes a pull request. That is a review step for the agent's intended work. There is no approval step that holds the agent until someone signs off, and no autonomy modes for low and high risk actions.

SourceSageOx, sageox.ai/docs plansread 2026-09-21

Every document carries a list of who may see it. Agents act as the person who connected them and see only what that person sees. Admins and members have distinct rights, admins write content rules that stop sensitive categories spreading, and API keys and user tokens are scoped. Hyperspell is SOC 2 certified by independent audit and GDPR compliant, and it says its policies are proven to third party auditors.

SourceHyperspell, docs.hyperspell.com resources/security, core/permissions and advanced/configuration, and hyperspell.comread 2026-09-21

Access control covers team invitations with owner, admin or member roles, MCP connections as scoped OAuth grants that see only the teams and repos the user belongs to and are listed and revocable in security settings, personal access tokens named in audit logs and revocable immediately, AES-256 encryption at rest, and redaction before observations are stored. SageOx publishes no certification or audit report.

SourceSageOx, sageox.ai/docs mcp, cli/pats, cli/team, discussions and your-dataread 2026-09-21

Every agent query to the brain is recorded, and each can be listed and retrieved with its answer and the sources behind it. Indexing progress is visible by provider. Hyperspell sees only the queries made to it, not the steps an agent took or the tools it called during a run.

SourceHyperspell, docs.hyperspell.com api-reference/evaluation/list-prior-queries and agents/overviewread 2026-09-21

When a coding agent's session ends, ox commits it to the repository's Ledger and summarizes it, covering the questions, the reasoning, the decisions and the code that came out of it. Sessions can be viewed one by one with ox session view, and each merged pull request links the session and plan that produced it. The Agent Toolkit also records stranded tool calls, and personal access tokens appear by name in audit logs. Together that is a record of what the agent did in each run, with audit logs beside it.

SourceSageOx, sageox.ai/docs session-recording, plans and cli/pats, and sageox.ai changelogread 2026-09-21

There are two isolated regions, the default United States region and a European Union region in AWS Europe (Ireland). The customer chooses a region at account creation, and all documents, embeddings and queries stay in it. Backups remain in the region, and data is never replicated across regions.

SourceHyperspell, docs.hyperspell.com resources/data-residencyread 2026-09-21

Team Context, Ledgers and recordings live on SageOx's hosted git server in us-west-2 with a local checkout synced to each machine, the CodeDB index stays on the customer's machine, and the open-source AI coworker is the customer's to run. The hosted data sits in one stated region, with no choice of region or self hosted option for the SageOx service.

SourceSageOx, sageox.ai/docs your-data and sageox.ai changelogread 2026-09-21

For agents a customer already runs, Hyperspell provides connectors, account-linking components and recommended system prompts. It names no catalog or set of ready-made agents or templates for a customer to adopt.

SourceHyperspell, docs.hyperspell.com integrations/overview and agents/overviewread 2026-09-21

The ox addons command installs curated addons (packaged skills, rules and context) into Team Context for every teammate. Three ship so far, among them the Agent Toolkit, a prebuilt hosted agent for Buzz and Slack that starts from the team's memory. ox coworker also lets a team add and load its own expert coworkers written as markdown. The addon catalog is documented in the changelog only.

Sourcesageox.ai/changelogread 2026-10-25

Customers choose which model answers queries. A fast default model is there for speed, and larger models are available for complex questions.

SourceHyperspell, docs.hyperspell.com resources/changelogread 2026-09-21

Each AI coworker file can set its model tier (opus, sonnet or haiku), a coworker's model can be overridden when it is loaded, the team can pick Codex or Claude as the brain behind a coworker, and repo-based coworker sessions run through the customer's own AI provider account across Claude Code, Codex, Gemini and other supported agents. The customer controls the model choice.

SourceSageOx, sageox.ai/docs cli/coworker, your-data and coding-agents, and sageox.ai changelogread 2026-09-21

A REST API with an OpenAPI specification lets developers add, update, query, list and delete memories, upload files, manage integrations, connections and folder policies, and issue user tokens. Hyperspell also offers an SDK, an MCP server for Claude, ChatGPT, Claude Code, Codex and Cursor, and a command line tool.

SourceHyperspell, docs.hyperspell.com api-reference/introduction, agents/mcp and agents/cliread 2026-09-21

A REST API with an OpenAPI description sits in SageOx's agent catalog beside an OAuth 2.0 protected MCP server whose tools search and read team knowledge and save sessions, agents can register through dynamic client registration, the ox CLI runs headless with personal access tokens for CI, cron and cloud agents, and developers can build a custom adapter to bring a new agent in. Together that is an API with a published description, an MCP server with write actions, and a fit into CI.

SourceSageOx, sageox.ai/.well-known/ard.json and sageox.ai/docs mcp, mcp/manual, cli/pats and developers/custom-adapterread 2026-09-21

An Evaluation API lets the customer list prior queries, retrieve their results and score a query result or an individual highlight, feeding back on the context served to its agents. The scores rate Hyperspell's retrieval, not the agent's behavior. Hyperspell names no test cases, harness or quality gate for the customer's agent.

SourceHyperspell, docs.hyperspell.com api-reference/evaluationread 2026-09-21

There is no testing of agents against fixtures or datasets, no scoring of output quality and no configurable quality gate. SageOx covers capture, Team Context, priming, session recording, plans and integrations.

SourceSageOx, sageox.ai/docsread 2026-09-21

The web crawler recursively crawls a site to make it available for indexed search, building the corpus ahead of time. Hyperspell names no browser, desktop or computer control for an agent at run time.

SourceHyperspell, docs.hyperspell.com integrations/all/web_crawlerread 2026-09-21

SageOx captures discussions through browser and desktop recorders, Ox Dot and imported audio or video, and there is no browser, desktop or computer control by an agent.

SourceSageOx, sageox.ai/docs context-captureread 2026-09-21