Pi, OpenCode, Codex, and Claude Code adapters implement one typed harness contract, while organization and scope configuration constrain approved harness and model choices.
Agent Atlas / Runtimes and workbenches
yc-software/qm
What is yc-software/qm, and what does it do?
A cloud-first organization agent runtime that gives people and shared rooms distinct scopes for sessions, memory, files, credentials, skills, schedules, and durable agent computers while routing several coding-agent harnesses through one core.
Organization agent runtime
Full dossier
Reported + sourced
Reviewed August 26, 2026
Intended operator
Startups self-hosting one agent system for individual and shared work across Slack and the web.
Primary surfaces
Slack / web workspace and admin panel / headless HTTP core / organization deployment CLI
Tracked repository
Teaching mechanism
The conversation address selects the tenant
A direct message resolves to a personal scope, a group message to a group scope, and a channel to a channel scope; the resolver then mounts organization state read-only and the active scope read-write before a harness runs.
Evidence limitThe project calls itself early and experimental; its security policy documents incomplete audience filtering, conditional egress enforcement, plaintext in-use credentials, durable-data retention gaps, and other controls that are not yet complete.
Inspect the source
System Map
Eight inspection boundariesPersonal, channel, group, team, and organization scopes separate writable and recalled memory, workspace layers, instructions, and session records.
A small core tool surface delegates commands and file effects to a scope-routed sandbox with local, Fly Sprite, and AWS MicroVM backends.
Organization policy composes with narrower scope policy; ACL handles, security postures, approvals, credentials, audience floors, and sandbox routing form separate enforcement layers.
Leased durable sessions, tapes, model-request records, background run stores, scheduler claims, and resident disks are designed for multi-instance and blue-green operation.
The stronger current evidence is shared human-agent tenancy and background work rather than a fixed hierarchy of role-playing agents.
Slack and web surfaces expose scoped conversation, approvals, admin policy, model selection, shared skills, schedules, and published app artifacts.
Public TypeScript interfaces, Postgres stores, threat model, deployment contract, tests, release workflows, and unusually detailed commit messages expose both mechanisms and known gaps.
Learn Through This Project
10 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 5
Delegation Is a Distributed System
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
Latest Reporting
6 articles
QM Lets Scheduled Scripts Finish the Job Before Calling an Agent
November 4, 2026 — The workplace agent’s new task-cron option runs a known procedure first. A handled result ends the scheduled run; exceptions carry the task and script output into the agent’s reasoning loop. yc-software/qm
Agents Need Contracts That Survive a HandoffOctober 27, 2026 — Katelyn Lesse’s engineering agenda is already visible in agent code. Five tracked projects offer starting points for three practical contributions: recover without duplicate effects, delegate without restoring old permissions, and stop without letting late work erase the stop. openai/codex, yc-software/qm, pydantic/pydantic-ai, Parcha-ai/agentrun, mastra-ai/mastra
QM Drops Two Agent Loops but Keeps ChatGPT AccessOctober 15, 2026 — The shared workplace agent is narrowing its runtime choices to Pi and Claude. Its October 14 code change shows why keeping a model connection does not require keeping the agent software that shares its name. yc-software/qm, different-ai/openwork
You Sent the Correction. Has the Agent Read It?September 27, 2026 — QM now places a mid-run correction where the agent runtime receives it. That small change separates a delivery problem from an agent that had the information and still went wrong. yc-software/qm
The Company Can Host the Agent. The AI Account Can Still Be Yours.August 31, 2026 — QM is separating the place an agent is operated from the account that pays for and authorizes one person’s model turn—while deliberately leaving background work on organization credentials. yc-software/qm
QM is trying to turn the personal agent into company infrastructureAugust 1, 2026 — Y Combinator's new open-source harness does not win its case by running Codex or Claude in Slack. Its more consequential bet is that every person and room needs a clearly owned memory, workspace, permission set, schedule, and durable computer. yc-software/qm, block/buzz, egregore-labs/egregore, openclaw/openclaw, NousResearch/hermes-agent, crewAIInc/crewAI, microsoft/agent-framework
Primary Sources
Open Questions
- Can scope ownership remain legible and enforceable as shared rooms accumulate credentials, memories, schedules, apps, and agents from several harnesses?
Profile facts were reviewed as of ; newer reporting is dated separately above. An evidence label is not a product score or proof that uninspected capabilities are absent.