Skip to content

Documentation index

Document Kind Purpose
Architecture Architecture Plan of record: what Humagen composes and the status table of what exists
The ecosystem Architecture The layers (Humagen and saige: its agent, flow, sandbox, durable) and the rules each keeps
One model, two surfaces Architecture How each agent concept reaches the web and Slack, and the open gaps
Projects Architecture Design of record for nested projects, policies, agents, inboxes and shared threads
Company, personal work and finding help Architecture The Company root project, personal projects, publishing work and Find Help
Conversation workspace Architecture The conversation page, activity and deliverables in the console
Session workbench Architecture The tabbed pane beside a session, how to add a tab, and how what a person did by hand reaches the agent
Agent calls Architecture How an answering agent calls other agents with the agent tool
Agent operations Architecture Agent controls, action policies, assignments and handoff contracts
Security Architecture Callers, roles, rules the code holds, credentials, inbox and break-glass reads
Sources Architecture Library sources from Google Drive, OneDrive and Notion, credentials and access
Library and personal settings Architecture Skills, Memory, Audit, Profile, Notifications, Privacy and Personal Agents
Managed Library and web foundation Architecture The console's shadcn foundation, theme, fonts and Library folder browsers
Native console notifications Architecture The Notifications page, badge and delivery preferences
Usage and Activity Architecture How Usage intervals and Activity classifications are computed
Egress guard Architecture How a refused write to Slack or GitHub is stopped, paused or backed off instead of repeated
Typed learning dependencies Architecture How skills and reviewed memory link back to the runs that produced them
Develop with Tilt Guide One tilt up for the API, the console with hot reload and Postgres, reusable by projects built on Humagen
Deploy Guide Run Humagen locally, with Compose, Helm, Flux or humagen deploy
GitHub and Slack trial Guide Set up a local trial with a GitHub installation and a Slack app
Work out a refund path Guide Two native agents on one conversation's booking records: one reads them and current policy pages and cannot write, one writes the approved request
Build a sourced chronology from records Guide A native agent reads the records in a conversation, cites and quotes each dated event, lists unknowns, and opens current official pages
Reconcile account exports Guide A native agent runs a shipped script over authorized exports in a sandbox, reports what does not reconcile, and drafts a plan whose amounts are all sourced
Research current public sources Guide A native agent with web_search and web_fetch cites the pages it opened
Background work Guide Watches, goals and tasks an agent keeps at while you do other things, and notify; you are told in your inbox when each ends
Schedules from a conversation Guide An agent proposes recurring work in a conversation or Slack thread, you approve it on a card, it confirms the next run, and running it once now tries it
Connect a Mac desktop Guide Pinned native MCP tools, permissions, targeting instructions, and an opt-in fixture check
Connect a browser Guide Pinned Playwright MCP configuration, existing-tab connection, and real browser acceptance
Read selected mail and approve a calendar change Guide A native agent on the person's own Gmail and Google Calendar connections, with each write approved in Humagen
Read mixed evidence Guide What a native agent reads of attached PDFs, images and audio, and the transcription path left after #383
Read a cloud document and approve an edit Guide A native agent on the person's own Google Docs and Drive connections, and what revision history depends on
Email notifications Guide Configure the outbound email notification relay
Upgrading from the old name Guide What changed in the rename and how an operator moves to the new names
Upgrading to saige v0.45.0 Guide What changes for operators and Go programs now that orchagent is folded into saige
Names Guide The ecosystem's current names and the retired ones
Rename verification Guide Verification of the install and upgrade paths after the rename
The Humagen spec Reference The System and Platform configuration documents
API contract Reference The OpenAPI contract, events, commands and the generated client
Authorization Reference The policy language: grammar, roles and grants
Clients Reference The client record for Slack apps and GitHub installations
Slack agent delivery Reference The Slack webhook connector and Slack app: configuration and behavior
GitHub agent delivery Reference The GitHub connector: mentions, labels, reviews and checks
Design requirements Reference Console design requirements (DES-NNN) and their implementation evidence
Workspace design principles Reference Principles for the console, with the reference each one applies
Design reference research Reference Products and people chosen as references for the working interface
Board design rules Reference Format, tokens, identity and navigation rules for the design boards
Control plane gap analysis Reference RFC 0001 checked against the code: built, partial, not built, diverges
Console UI review Reference Console UI review findings, patterns and how to check a UI change
RFC 0001: Agentic workspace and control plane RFC, Mostly built Product and design spec for delegated agent work; design notes in assets/control-plane
RFC 0002: Cross-project handoffs RFC, Proposed A staged design for handoffs that cross project boundaries
RFC 0003: Conversation-first delivery RFC, Partly built Conversation-first delivery and collections for follow-on work
RFC 0004: Isolated runs and person context RFC, Mostly built One agent for many people in parallel: pinned person context, isolated sandboxes, and the path to a million people in one room
RFC 0005: Federated retrieval and the catalog RFC, Mostly built Live search in external systems, a metadata catalog, and hybrid search for what Humagen owns
RFC 0006: Multiple replicas, fiber folded in, and a durable event bus RFC, Partly built fiber's fold completed, the standalone repository deprecated, and the Postgres bus with an outbox and consumer groups for many replicas
RFC 0007: Agent marketplace and credits RFC, Not planned A hosted Humagen with credits and a paid agent marketplace, kept as a record
RFC 0008: Feedback, run signals and training export RFC, Partly built Asking for ratings at the right moment, per-run signals and reward analytics, and consented, redacted KTO and DPO export
RFC 0009: One entry point RFC, Proposed One agent, named for the platform, routes each ask to the specialists and answers once
RFC 0010: Corpora and the workspace graph RFC, Proposed Managed knowledge, search, tools and ownership; supersedes the corpus model of RFC 0003
RFC 0011: Handoffs and threads RFC, Proposed Who answers, how agents reach each other by tool call, and where their work shows
Roadmap Plan Product direction by area and the next delivery boundary
Task queue Plan Current UI delivery and follow-on platform work, in order
Workspace implementation plan Plan Tasks and requirement traceability for the workspace and control plane
Implementation constraints Plan Constraints resolved before implementing the workspace plan
Prototype control audit Plan Acceptance status of every interactive control in the prototypes
Final native acceptance Plan Recorded checks for the implemented workspace and control plane
Local browser validation Plan Incremental browser evidence from the persistent local preview
Application validation Plan Local validation of the workspace prototype's application work

Design assets: assets/control-plane (design notes; the clickable boards were removed) and assets/workspace-prototype (interactive prototype and sample files). Requirement and audit records for the implementation plan are in assets/workspace-implementation/.