SLIM WG Summary - June 23rd 2026

[NickG44]

SLIM Working Group Call Summary – June 23, 2026

Key Topics & Outcomes

1. Control Plane & Peer Auto-Discovery Demo

• A live demonstration showcased the new architecture dividing routing responsibilities:
  • Intra-Group (Data Plane): Nodes within the same deployment auto-discover each other (via static lists or Kubernetes APIs), form a full mesh, and distribute subscriptions locally without the control plane.
  • Inter-Group (Control Plane): The control plane exclusively manages routing between different deployments/groups using Shortest Path Trees to prevent network loops.
• Topologies & Segmentation: The controller can enforce specific inter-group topologies (e.g., chains, full mesh, or dynamic stars). A new "Segmentation" feature was demoed, allowing strict network isolation (e.g., multi-tenant environments) where groups in different segments cannot see or route to each other.
• Fault Tolerance: The demo proved that if a gateway node dies, the controller automatically detects the failure, selects a new gateway node, and recalculates the routes seamlessly.

2. Workspace Refactoring & Benchmarking

• Merged: The massive PR to refactor the core repository into a single, flat Rust workspace (crates/) has been merged. Task files are now at the root level, and CI caching is significantly optimized.
• Rebasing Warning: Contributors with open PRs must carefully rebase, as Git may misinterpret moved files as deleted.
• Benchmarking Regressions: Recent improvements to the SlimCTL bench tool uncovered some regressions in the "unreliable" mode for group channels. The team will prioritize updating end-to-end tests to catch these edge cases and plans to initiate a brief code freeze soon for comprehensive testing.

3. WebAssembly (Wasm) & Header Integrity

• Wasm Data Plane Merged: The PR to compile the entire SLIM data plane to WebAssembly has been successfully merged. The final remaining step for full browser support is implementing browser-native WebSockets (using gloo-net).
• Header Integrity: The PR adding cryptographic integrity checks to packet headers (to detect tampering) is fully refactored to support Wasm and is ready to merge in the coming days.

4. Ecosystem Integrations (MCP & A2A)

• Python MCP Bindings: A community member noted that the slim-mcp-python repository is using an outdated version of the SLIM bindings (v1.1), causing integration issues with custom MCP servers. The team will immediately bump this dependency to v1.4.1 and look into automating future bumps via Renovate.
• A2A WebSockets: The architectural debate continues regarding the A2A WebSocket custom protocol binding. The A2A TSC is questioning whether WebSockets should be a standalone, first-class transport for unary messages, or simply an upgrade path from HTTP/2 strictly for bidirectional streaming.

5. Shadi CLI & Multi-Agent Cooperation

• Development continues on the "Agent Bridge" feature within the Shady CLI, allowing local CLI agents to securely share context and cooperate over SLIM.
• The team is testing whether orchestrating multiple smaller, local models (e.g., 4 models with 250k context windows) can effectively replace a single massive, expensive model (1M context window) for complex tasks.

---

Next Steps & Action Items

• Controller Release: Merge the pending segmentation PR, update the corresponding Helm charts, and cut a new release to allow for comprehensive multi-cluster testing.
• End-to-End Testing: Update the E2E test suite to cover the regressions found in the unreliable group channel mode.
• Browser WebSockets: Begin work on the final PR to implement browser-native WebSockets (gloo-net) for the Wasm target.
• MCP Dependency Bump: Update the SLIM bindings dependency in the slim-mcp-python repository to v1.4.1 to unblock community MCP server development.
• Header Integrity: Merge the finalized header integrity PR into the main branch.