Orchestration exploration - #172
Draft
mogul wants to merge 5 commits into
Draft
Conversation
I found biomelab and I think it seems well-suited, but I don't know anything about the usability
1. L2 generalized — diagram now reads "Isolation / execution boundary" with SBX as default/strongest and container/Unix-user/worktree-only as alternatives; cross-reference added. 2. New section "Native Isolation, Independent of SBX" — strength spectrum (microVM → container → Unix-user+sudo → worktree/none), per-tool mechanism table, the two real alternatives (vigilante planned container, agor Unix-user) with caveats, and an explicit Lima/nono not found callout. 3. agor local-vs-shared-server subsection — confirms it runs locally in simple mode (no sudo/Postgres) but gives no isolation there; isolation only via the privileged server model; daemon/MCP surface + BSL 1.1 apply in either mode. 4. Comparison table — replaced "SBX/sandbox-aware" with a verified "Native isolation tech" column; agor row updated to "Local (simple) or shared server (RBAC)". 5. Where each tool sits — kept the L4/L5 split (Option B) and added a deployment-model axis (local single-dev ↔ shared-server multi-user). 6. Constraints/open questions — added the no-privilege-escalation conflict for agor and a new L2-alternatives question (Lima/nono/vigilante container). 7. Metadata — OpenChamber URL → btriapitsyn/openchamber; added isolation-source references.
- OpenShell added as an L2 runtime — surveyed as a dedicated SBX peer (microVM/container drivers, YAML policy across FS/net/process/inference, L7 egress proxy, Privacy/inference router); added to the isolation spectrum, native-isolation table, main comparison table, and a new "L2 isolation runtimes" grouping. - Agent of Empires (AoE) added as a co-equal orchestrator — across the native-isolation table, comparison table, agent-agnostic note, and the L5/deployment-model lists (native Docker sandbox; --cmd-override shim seam). - New "Shimming SBX into Non-Isolating Tools" section — four seam types: (a) explicit command/wrapper config, (b) PATH override, (c) server-API/no-seam, (d) orchestrator-level runtime template; with a worked shim sketch, a per-tool best-seam table, and the four hard problems (worktree mount, qsbx two values, PTY passthrough, federal trust caveat). - agor reframed around its executor template — added the #agor-executor-template subsection; verified against shipping source (main 21edf7b) that executor_command_template is implemented (not just an exploration doc), with the corrected/actual template-variable list and the docs-vs-code drift noted. - agor BSL 1.1 de-flagged — corrected throughout to "self-host OK; only restricts reselling as a hosted service," per your input. - Generalized L2 in the layer model — diagram + framing changed from hard-coded "SBX sandbox" to "isolation/execution boundary (SBX default)" with the alternatives spectrum. - Supporting updates — sharpened the BiomeLab mount open-question (resolves to --kit or qsbx-prepares-sandbox, not mounts), expanded open questions (L2 alternatives incl. OpenShell/nono/Lima; shim strategy; agor integration), and added AoE/OpenShell/nono/seam-source references.
- SBX --clone abandoned — L1 section retitled and rewritten around git-worktrees-mounted-into-sandboxes; the comparison table reframed as chosen/abandoned with the data-loss footgun as the stated reason; conceptual-layers diagram updated (L1 = "git worktree (mounted into the sandbox)"). - agor split into three rows in both the native-isolation and main comparison tables (simple / insulated-strict / executor-template), plus a callout clarifying they're one product in different configs. - Pilot Candidates section — generalized from BiomeLab-only to two candidates (BiomeLab + agor), and moved to sit just before Constraints & Open Questions. - agor worktree-mount reframed as largely-solved-by-design (agor manages the worktree on shared data_home before the executor runs; sandbox just mounts data_home) — in the executor-template section, the pilot, and open question 6. - Shimming section de-emphasized — added a "too fragile" callout pointing to agor's executor template as the durable seam. - Open questions rewritten per your direction: dropped the side-question parenthetical (Q1); L1 = standardize on worktrees (Q2); added re_gent as a governance candidate (Q3); SBX-not-settled / evaluate OpenShell / exclude nono (Q4); shim de-emphasized (Q5). - OpenShell baseURL speculation removed (USAi is OpenAI-shaped) across the narrative bullet, table, and open questions. - Stray </content>/</invoke> tags removed from the end of the file.
Contributor
Author
|
Here's a more-focused 15-minute video that shows Agor off pretty well. |
Adds a transcript of a session with USAi about how to compare three L4 orchestration options against the federal solo or small-team use-case: Omnigent (from Databricks), Agor, and BiomeLab
Contributor
Author
|
@rpalmer-gsa I added that rumination session I mentioned; you probably want to ingest that into whatever you were using to make your diagram earlier today. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Documents some of my meandering exploration of viable options for "agent-farming" (usign a single-pane of glass and orchestrator across many agent instances) in the federal context
Related Issues
None
Type of Change
Note this is a draft... I don't expect it to be merged, and am just making the PR for visibility and feedback.
Checklist
feat:,fix:,docs:)Test Results
N/A
Additional Notes
Again, this draft PR may never get merged; I'm just using the PR mechanism for visibility. It's possible that this belongs in the -patterns repo; I just haven't looked at that one at all yet.
Y'all, agor looks amazeballs. Check out this demonstration (from February! lots was added since then) and have your mind 🤯!