swarm-rd-orchestrator
About
Append-only event log for cross-agent context/memory sharing via 3 MCP tools.
Details
- Author
- rudrendupaul
- Categories
- AI
Jump to
Setup
Install swarm-rd-orchestrator in your MCP client (Claude Desktop, Cursor, Windsurf, and others).
Repository: https://github.com/rudrendupaul/swarm-rd-orchestrator
Follow the installation instructions in the repository README, then restart your MCP client.
What is swarm-rd-orchestrator-cli, and why does it exist
swarm-rd-orchestrator-cliis a shared, durable event log for parallel AI research agents built on Ray's actor model. Each agent writes findings as structured deltas; any other agent can pull the full history for a task without a shared mutable store or a coordinating server process.
It exists to test a specific, narrow hypothesis: that Ray's actor and object-store model is a better fit for coordinating genuinely large numbers of parallel research agents than an HTTP-based coordination layer. That hypothesis is unproven. The project ships as a Milestone 1 spike specifically to test it against a real workload before any further investment, seeRun the actual validation spike.
git clone https://github.com/RudrenduPaul/swarm-rd-orchestrator.git cd swarm-rd-orchestrator python3 -m venv .venv .venv/bin/pip install -e ".[dev,mcp]"
Requires Python 3.10 or newer (needed for themcpSDK dependency).
9/9 passing, including the load-bearing test: 3 concurrent Ray actors appending to one shared event log, reconciled with zero lost or duplicated deltas.
EditREAL_TASK_ID,REAL_TASK_DESCRIPTION, and the sample agent findings inspike.pyto reflect a real research task, then:
Read the printed rubric at the end. The decision rule: fewer than 3 qualifying architectural failure cases against raw Ray or LangGraph means falling back to a thin CLI wrapper instead of building this out further.
- Primitive: Ray (actor model and object store), LangGraph as fallback
- Storage: SQLite, WAL mode
- Delta shape:{task_id, agent_id, timestamp, content, kind}
- Malformed delta raisesInvalidDeltaError, never silent
- License: Apache 2.0
- Python: 3.10 or newer, required for themcpSDK
- Publishing: live onPyPI(via GitHub Actions OIDC Trusted Publishing, no stored token) andnpm, both verified with a real clean-environment install
What does this actually do?It gives parallel AI agents, running as Ray actors, a shared and durable place to write findings and pull each other's, without stepping on each other's state. It's a transport and persistence layer, not an orchestration framework: it doesn't schedule agents or decide what they do.
How is this different fromswarmmesh?SeeComparisonabove. Same core idea, different transport (Ray actors instead of HTTP), andswarmmeshis currently the more complete, already-published option.
Does this work on Windows, macOS, and Linux?Tested on macOS. Ray itself supports Linux and macOS natively; Windows support for Ray is more limited upstream, so treat Windows as unverified for this project specifically until someone confirms it.
Does this need my own API keys?No. This project makes no LLM calls of its own. It's a coordination layer that your own agents, whatever model or framework they use, write to and read from.
Is this safe to depend on?No, not yet. This is a pre-validation Milestone 1 spike (versions0.0.x), not a stable release. The event log's core guarantees (atomic append, no partial writes) are tested intest_event_log.py, but the project hasn't been validated against a real multi-agent workload beyond the spike script yet.
How do I use this from an agent, not just a human?Either shell out to the CLI with--jsonon every command, or runswarm-rd-cli mcpand connect to it as an MCP server. Both give structured, parseable output.
Is this a library or just a CLI?Both.event_log.py'sEventLogandEventLogActorclasses are directly importable if you're already in Python and don't need the CLI or MCP layer.
What license is this under, and can I use it commercially?Apache 2.0. Commercial use, modification, and redistribution are all permitted under its terms; seeLICENSEfor the full text.
There's noCONTRIBUTING.mdyet since this is a pre-validation spike, not an accepted-contributions project. Open an issue if you want to discuss a change before this exists formally.
Apache 2.0. SeeLICENSEfor the full text.
Local-first agent memory: a plain-Markdown Obsidian vault is the source of truth, with a rebuildable DuckDB index for hybrid BM25 + vector + graph recall.
Persistent memory and semantic search for AI coding assistants across sessions
Give your agent a memory: shared, cited, tenant-isolated knowledge-graph memory for any MCP host. Grounded answers from a local-first June endpoint — abstains rather than guesses.
Decentralized persistent memory for AI agents — encrypted vault storage built on Walrus and Sui.
Persistent memory for AI assistants and coding agents across ChatGPT, Claude, Cursor, and other MCP-compatible tools.
Your portable AI memory vault — memories, skills & configs, shared across every AI tool.
Local Work Model for AI agents that learns from real outcomes.
Adaptive MCP memory system for AI applications. Learns which retrieval strategies work for your data, scores results using cognitive science models, builds a knowledge graph automatically, and validates every parameter change against real query history before adopting it. Patent pending.
Auditable, self-improving knowledge & memory for AI agents over MCP — citation-enforced answers and a replayable why-trace, self-hosted on Postgres.
Turns your task manager into agent memory: hybrid (RRF) retrieval over TickTick or an Obsidian vault via an adapter contract. MCP server + CLI, no vector DB to maintain.
Sign in to leave a review
Use Google, GitHub, or an email account so ratings stay tied to real people.
No reviews posted yet.
