Agent Atlas / Runtimes and workbenches

different-ai/openwork

An OpenCode-powered workbench that turns team memory, self-hosting, desktop packaging, authentication, MCP sessions, and shared workspaces into operator-visible product boundaries.

Agent workbench Scope profile Reported + sourced Reviewed August 26, 2026

Intended operator

Teams hosting and sharing coding-agent workspaces.

Primary surfaces

desktop workbench / server and self-hosting / shared workspace

Tracked repository

different-ai/openwork

Teaching mechanism

Authentication needs a resumable state, not a timeout

A sign-in flow remains pending with a claim link and explicit state instead of turning an operator action into an arbitrary elapsed-time failure.

Evidence limitA clearer state machine does not prove an identity provider, deployment, or client will always deliver the callback.

Inspect the source

System map status

Reporting is still in progress.

This scope profile names the project's place in the field and one useful mechanism. It does not claim that every runtime boundary has been inspected.

Learn Through This Project

9 lessons

Chapter 1 Before the First Model Request Chapter 2 The Invisible Prompt Chapter 3 Tool Calls: Contract, Authorization, Execution Chapter 4 Loops, Workflows, and Recovery Chapter 6 Memory Is a Typed Storage Contract Chapter 7 The Approval Trap Chapter 8 The Human Control Loop Chapter 9 Trust Must Make Its Case Chapter 10 The Future Has Three Branches

Compare Its Boundaries

7 labs

Comparison labTool Contracts Comparison labGoverned Memory Comparison labApproval Boundaries Comparison labHuman Workbench Comparison labSessions And Recovery Comparison labEvidence And Evaluation Comparison labMaturity Signals

Latest Reporting

8 articles

A Child Agent’s Approval Request Shouldn’t Hide in Another Tab

September 2, 2026 — Openwork now brings a child session’s blocked permission into the parent task; fresh Goose state-machine work shows why seeing the ask is only half the control. different-ai/openwork, aaif-goose/goose

The Last Click Shouldn’t Redirect an Automation

August 30, 2026 — OpenWork has stopped a scheduled Automation from silently following whichever workspace a person activated last; a job can now carry its intended place with it. different-ai/openwork

OpenWork learns to wait for a turn that has actually started

August 8, 2026 — OpenWork’s new headless-thread client turns a subtle race into an explicit contract: an agent turn is not finished just because its status looks idle. different-ai/openwork

An Agent's Sign-In Link Is Not a Timeout

July 21, 2026 — OpenWork's newest MCP work carries a provider's structured authorization link through a failed tool call, so the person who owns the account can act instead of watching an agent retry the wrong problem. different-ai/openwork

OpenWork Gives MCP Failures a Useful Address

July 13, 2026 — A new diagnostics layer separates DNS, TLS, HTTP, OAuth, protocol, and provider failures, then returns a bounded reference and the person who can act—without logging the raw connection secret trail. different-ai/openwork

Harness Engineering Is Becoming The Field's Self-Improvement Layer

July 8, 2026 — Lilian Weng's July 4 map of self-improving harnesses gives builders a constructive middle path: improve the runtime around the model, but keep evidence, permissions, and evaluation legible enough that the loop can be trusted. lilianweng/lilianweng.github.io, openai/codex, different-ai/openwork, openclaw/openclaw, NousResearch/hermes-agent, earendil-works/pi, anomalyco/opencode

Self-Evolving Agents Need Receipts For The Loop

July 8, 2026 — Shilong Liu's taxonomy widens the self-improvement story beyond harnesses: agents may optimize artifacts, modify the harness, or learn in the model, but each loop needs evidence about what changed and what stayed outside the optimizer. NousResearch/hermes-agent openai/codex openclaw/openclaw different-ai/openwork anomalyco/opencode

OpenWork Turns OpenCode Into A Shareable Agent Host

July 6, 2026 — OpenWork's source trail shows a desktop app, server, orchestrator, router, file sessions, approval flow, export safety, and single-org deployment policy around OpenCode rather than a thin UI wrapper. different-ai/openwork, anomalyco/opencode

Primary Sources

OpenWork repository

Primary workbench and self-hosting source.

Open Questions

Can a team see who owns saved context, live credentials, and every shared session effect?

Evidence state describes what this Atlas profile can support today. It is not a product score and it is not evidence that uninspected capabilities are absent.