Skip to content

dsivov/cohermes

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

16,119 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

cohermes — a team of AI coding agents that share one brain

cohermes

A team of AI coding agents that share one brain.

cohermes is an extension of Hermes that swaps its per-user memory for a shared Context Graph.

Hermes gives each developer a capable, autonomous personal agent — with memory that is private to that user. cohermes points a whole team's agents at one shared Context Graph: every agent reads project context from the graph and writes its decisions, code reviews, tasks, and commits back into it. The team reasons over one living project instead of N amnesiac silos.

Hermes remembers, per developer. cohermes makes the team remember — together.

What cohermes adds

Hermes cohermes
Memory per-user, local (MEMORY.md) team-shared Context Graph (governed, queryable)
Knowledge one operator's recall decisions · code reviews · tasks · commits, cross-linked
Reuse within your own sessions a teammate's why is reused, not re-derived
Sync pull-only, via Hermes cron (no push bus needed)
Cost your choice subscription-only, enforced (never metered API)

Everything Hermes does — its runtime, 40+ tools, terminal backends, orchestration — is inherited unchanged. cohermes is the cg/ package on top.

How it works

  • One project = one CG workspace, selected by a header; fully isolated.
  • The loop, per developer: orient → query-before-build → work → record-the-why, through the context-graph MCP tools.
  • The chain: Decision –motivates→ Task –implemented_by→ Commit –reviewed_in→ Review –enacts→ Decision, all traceable.
  • Sync: a cron watcher polls the graph and tells your agent what teammates changed.
  • Auth: runs on your Claude subscription — a preflight guard refuses metered API billing.

Quick start

# point at your Context Graph server + your project's workspace
export COHERMES_CG_URL=http://<cg-host>:9621
export COHERMES_CG_WORKSPACE=<your-project>

python -m cg.ontology            # install the shared 4-artifact ontology (once)
python -m cg.preset ./my-repo    # wire a repo: .mcp.json + CLAUDE.md loop + slash-commands
python -m cg.auth                # assert Claude runs on your subscription

See cg/README.md for module details. The full design log and decision record live in the companion Context Graph repo (docs/TEAM_AGENT_FRAMEWORK.html).

Status

Walking-skeleton MVP: the full chain records, cross-links, and traces; a teammate's agent reuses prior decisions via precedent search instead of re-deriving them — measured against a fresh control workspace.

Credits & license

cohermes is a fork of NousResearch/hermes-agent (MIT). The Hermes runtime, tools, and orchestration are the work of Nous Research; the cg/ extension is cohermes. The original Hermes README is preserved as README.hermes.md, and the upstream MIT license (© 2025 Nous Research) is retained unchanged in LICENSE.

Licensed under the MIT License.

About

A team of AI coding agents that share one brain — a Hermes extension whose per-user memory becomes a shared, governed Context Graph.

Topics

Resources

Contributing

Security policy

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages