Deployment shapes: machine-as-brain vs mesh-brain lobe profiles — spec + plan (0.41.1) by OriNachum · Pull Request #113 · agentculture/lobes-cli

OriNachum · 2026-07-14T04:41:48Z

Spec + plan (devague /think → /spec-to-plan, both frames converged and committed): deployment shape becomes a first-class, safely-selectable axis — orthogonal to the per-machine hardware profiles that landed in #110.

The two shapes

machine-as-brain (default, unchanged behaviour): one box hosts every lobe it can serve. Not everyone has many machines — the one-box path stays first-class and zero-decision (bare lobes init, goldens byte-identical).
mesh-brain lobe profiles (opt-in, per box): each box hosts the lobes it is best at. Near-term shapes: spark-lobe keeps the Qwen cortex + embedder + reranker + stt/tts and drops Gemma senses; thor-lobe keeps Gemma senses + the rest and drops the Qwen cortex.

Selection is flag-first: lobes init --shape <machine-as-brain|spark-lobe|thor-lobe>, dry-run by default, --apply to commit (wizard parked as follow-up). Cross-box story for this change is per-box honesty only; gateway proxying and the brain-level view are deferred to the #112 end-state design.

Aligned to the shipped #110 substrate (updated mid-flight)

#110 (the #108 implementation) merged while this PR was open; the branch is merged with main (0.41.0 → this PR bumps 0.41.1) and the spec/plan were re-cut against what actually shipped:

Dropped-lobe contract = flag, not omit: capabilities show the role feasible:false (annotated), /v1/models omits it, and the generate lane returns 404 role_infeasible — targeting the landed machinery (lobes/gateway/_config.py FEASIBLE_ENV, _routing.py, server.py::_feasibility_response, roles.py), never a parallel path.
Shapes own all six Colleague roles (user decision): stt/tts are first-class shape members — the four core roles render via the landed Profile machinery (lobes/profiles/schema.py / loader.py / render.py), the audio pair via the audio overlay.
Shape data files are TOML, matching lobes/profiles/builtin/; per-(shape,card) goldens are additive to the shipped tests/goldens/*.env + regen.py convention.
The plan's recorded risk "the spec+plan: lobes fits the machine it lands on — per-machine hardware profiles #108 implementation hasn't landed yet" is resolved by feat: per-machine hardware profiles — detect the card, fit the box (13-task plan) #110 — the alignment sweep verified all six assumed surfaces exist under the assumed names.

Evidence this stands on

verify on the GB10 before t3 ships: is the fp8-KV crash checkpoint-driven, and is VLLM_ATTENTION_BACKEND truly dead on the pinned image? #109 verified and closed (read-only, from the live GB10 deployment): fp8-KV is an sm_110-conditional trait (the same pinned nightly + scale-less checkpoint boots fp8 on the GB10 and only warns), and VLLM_ATTENTION_BACKEND is dead on the GB10's pinned image too (undeclared in vllm.envs; TRITON_ATTN comes from Gemma4 model-side forcing).
Before-state is checkable: the four core roles carry no profiles: stanza in the fleet template — today "Spark drops Gemma" means hand-editing the compose.
Co-residency tax (the why): senses trimmed to 32K, cortex to 128K vs 256K solo-native, inside the 0.56 GB10 budget.

The plan: 8 tasks, 5 waves

t1 shape schema → (t2 budget re-derivation | t3 render+goldens | t5 dropped-lobe honesty) → t4 init --shape → (t6 GB10 validation | t7 Thor validation) → t8 docs+evidence. Wave-1 tasks are file-disjoint; every task carries TDD acceptance criteria and OWNS-scoped instructions. Remaining risks: t7 needs hands on the physical Thor; reclaim numbers are measured in t6, not specced.

Related: #108/#110 (substrate, landed), #109 (closed by this verification), #112 (one-lobe-per-box end-state, filed as part of this work), #105/#106 (rerank ordering, expected-failure gate in t7).

Version: 0.41.1 (every-PR-bumps rule; merged with main's 0.41.0). Docs/spec/plan only — no source changes; 1419 tests pass, agex pr lint clean.

…esh-brain (devague /think) Deployment shape becomes a first-class, safely-selectable axis orthogonal to the #108 hardware profiles: machine-as-brain (default, one box hosts every lobe it can) vs per-box mesh-brain lobe profiles (spark-lobe drops Gemma senses; thor-lobe drops the Qwen cortex). Cross-box story: per-box honesty only; one-lobe-per-box end-state tracked as #112. Grounded by the #109 verification (closed): fp8-KV is an sm_110-conditional trait and the attention-backend env is dead on the shared pinned nightly. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01U9ybFURhQxT9UpCgQNVL67

t1 shape schema -> (t2 budgets | t3 render+goldens | t5 dropped-lobe honesty) -> t4 init --shape -> (t6 GB10 validation | t7 Thor validation) -> t8 docs. Version 0.40.4 (every-PR-bumps rule). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01U9ybFURhQxT9UpCgQNVL67

OriNachum · 2026-07-14T04:41:50Z

/agentic_review

qodo-code-review · 2026-07-14T04:41:58Z

Code Review by Qodo

New Review Started

This review has been superseded by a new analysis

qodo-code-review · 2026-07-14T04:42:40Z

PR Summary by Qodo

Add deployment-shapes spec + implementation plan (0.40.4)

📝 Documentation ⚙️ Configuration changes 🕐 20-40 Minutes

AI Description

• Add spec for deployment shapes: machine-as-brain default vs spark/thor mesh-lobes.
• Add 8-task implementation plan with waves, acceptance criteria, and scoped ownership.
• Bump version to 0.40.4 and record the change in CHANGELOG.

Diagram

graph TD
  A[".devague/current + current_plan"] --> B["Devague exports (frame + plan JSON)"] --> C["Docs: spec + plan (markdown)"] --> D["Release metadata: CHANGELOG.md + pyproject.toml"]

High-Level Assessment

The following are alternative approaches to this PR:

1. Markdown-only spec/plan (no devague JSON exports)

➕ Reduces repository churn/size from large exported JSON blobs
➕ Keeps review surface focused on the published docs
➖ Loses structured data that may drive tooling/workflows around frames/plans
➖ May diverge from established devague process expectations

2. Track the 8 tasks as GitHub issues/milestone instead of an in-repo plan

➕ Task execution and discussion live where implementation happens (issues/PRs)
➕ Easier progress tracking and assignment across operators
➖ Harder to enforce “single source of truth” for acceptance criteria/OWNS scopes
➖ Plan may fragment across multiple issues without a cohesive narrative

Recommendation: Keep the current approach if devague exports are part of the team’s normal workflow (the JSON provides structured, auditable claims/tasks), but consider whether both JSON + markdown are necessary long-term. If reviewers find the JSON exports noisy, a follow-up could generate markdown from JSON in CI and only commit the markdown.

Files changed (8) +805 / -3

Documentation (7) +804 / -2

currentPoint devague 'current' to the new brain-shapes frame +1/-1

Point devague 'current' to the new brain-shapes frame
• Updates the current devague frame slug to the new deployment-shapes spec.
.devague/current

current_planPoint devague 'current_plan' to the new brain-shapes plan +1/-1

Point devague 'current_plan' to the new brain-shapes plan
• Updates the current devague plan slug to the new deployment-shapes plan.
.devague/current_plan

lobes-serves-the-brain-shape-you-choose-machine-as.jsonAdd devague frame: deployment shapes spec (machine-as-brain vs mesh-lobes) +304/-0

Add devague frame: deployment shapes spec (machine-as-brain vs mesh-lobes)
• Introduces the exported devague frame capturing the spec: default machine-as-brain vs opt-in mesh-brain shapes (spark-lobe/thor-lobe), plus requirements around safe selection, per-box honesty, and budget reclaim.
.devague/frames/lobes-serves-the-brain-shape-you-choose-machine-as.json

lobes-serves-the-brain-shape-you-choose-machine-as.jsonAdd devague plan: 8 tasks in 5 waves for brain-shapes implementation +326/-0

Add devague plan: 8 tasks in 5 waves for brain-shapes implementation
• Adds the exported devague plan with task decomposition (t1–t8), acceptance criteria, dependencies, OWNS-scoped instructions, and risk tracking for implementing shape selection and the initial spark/thor shapes on top of the #108 substrate.
.devague/plans/lobes-serves-the-brain-shape-you-choose-machine-as.json

CHANGELOG.mdAdd 0.40.4 changelog entry for deployment-shapes spec + plan +6/-0

Add 0.40.4 changelog entry for deployment-shapes spec + plan
• Records a new 0.40.4 release entry describing the added spec/plan for deployment shapes and the intended CLI shape selection and honesty behavior.
CHANGELOG.md

2026-07-14-lobes-serves-the-brain-shape-you-choose-machine-as.mdPublish markdown build plan for deployment shapes +87/-0

Publish markdown build plan for deployment shapes
• Adds a human-readable plan document enumerating tasks, ownership boundaries, acceptance criteria, and risks for implementing deployment shape selection and validation.
docs/plans/2026-07-14-lobes-serves-the-brain-shape-you-choose-machine-as.md

2026-07-14-lobes-serves-the-brain-shape-you-choose-machine-as.mdPublish markdown spec for deployment shapes +79/-0

Publish markdown spec for deployment shapes
• Adds the spec describing the two shapes, why they exist, constraints (orthogonal to hardware profiles), safety requirements (dry-run/--apply), and scope boundaries (no cross-box proxying; end-state tracked separately).
docs/specs/2026-07-14-lobes-serves-the-brain-shape-you-choose-machine-as.md

Other (1) +1 / -1

pyproject.tomlBump project version to 0.40.4 +1/-1

Bump project version to 0.40.4
• Updates the package version from 0.40.3 to 0.40.4 to satisfy the project’s per-PR bump rule.
pyproject.toml

qodo-code-review · 2026-07-14T04:44:06Z

Code Review by Qodo

🐞 Bugs (0) 📘 Rule violations (0) 📎 Requirement gaps (0)

Great, no issues found!

Qodo reviewed your code and found no material issues that require review

…spec/brain-shapes Resolves the every-PR-bump collision on CHANGELOG.md / pyproject.toml / uv.lock by taking main's side (0.41.0); the branch's own 0.40.4 bump is dropped and re-applied on top as 0.41.1. #110 landed the #108 profile substrate this spec+plan composes on. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01U9ybFURhQxT9UpCgQNVL67

…erged base) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01U9ybFURhQxT9UpCgQNVL67

#110 merged mid-flight (the #108 implementation, 0.41.0), so the frame and plan are re-cut against what actually shipped: the dropped-lobe contract is now flag-not-omit (capabilities feasible:false, /v1/models omits, generate lane 404 role_infeasible — c11/h3 rejected, c17/h13 recaptured and covered by t5); shapes own all six Colleague roles per user decision (stt/tts are first-class shape members over the audio overlay — c18, t1); shape data files are TOML matching lobes/profiles/builtin/; t1/t3/t5 instructions now name the landed modules instead of 'when it lands'. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01U9ybFURhQxT9UpCgQNVL67

sonarqubecloud · 2026-07-14T05:12:39Z

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

…nses (#113) With the co-resident heavy lobe gone, the remaining heavy lobe gets its FULL native budget in this PR rather than deferred to #112: - spark-lobe [overrides.cortex]: gpu_mem_util 0.44 -> 0.60 (the proven GB10 primary value, not the co-resident reclaim-sum), max_model_len 131072 (no override, i.e. co-resident) -> 262144 (the checkpoint's native 256K). - thor-lobe [overrides.senses]: gpu_mem_util stays 0.44, max_model_len 32768 (co-resident) -> 131072 (the Gemma checkpoint's native 128K). tests/test_shape_budgets.py assertions updated first (TDD) to demand the new values; goldens regenerated via tests/goldens/regen.py -- only the shapes/ goldens affected by these two overrides changed (thor-lobe__base.env and the four flat profile goldens are byte-identical, since senses is infeasible on the base card and the card baselines are untouched). Known follow-up: tests/test_init_shape.py hardcodes spark-lobe PRIMARY_GPU_MEM_UTIL as "0.44" in three places (lines 132, 166, 228) and now fails post-rise to 0.60 -- out of this task's owned-files scope (any Python source besides tests/test_shape_budgets.py), left for a follow-up.

Live validation caught the gap: `lobes init --shape spark-lobe --apply` rendered MULTIMODAL_FEASIBLE=false honestly, but the base compose kept all four core services unconditional, so `lobes fleet up` still booted the dropped senses lobe (observed: model-gear-vllm-multimodal Started) and ate the exact GPU budget the shape reclaimed. A dropped lobe must not RUN, not merely be flagged. Fix — a GENERATED compose override, mirroring the audio-overlay pattern: - `lobes init --shape <mesh-shape> --apply` now writes docker-compose.shape.yml (only when the shape drops >=1 core role) that parks each dropped core service in the inert `shape-dropped` compose profile (nothing activates it, so `docker compose up` skips it) and clears the gateway's depends_on with the compose `!reset` tag (removes the dangling edge to the profile-disabled service; the remaining default gears start regardless of ordering, which the gateway tolerates). Content is derived from the t3 render API (shape_render.ROLE_SERVICE over schema.ROLES), not hardcoded per shape. machine-as-brain / bare init write nothing (byte-identical scaffold); re-init to machine-as-brain scrubs a stale override; dry-run reports write/remove. - _compose.py gains SHAPE_OVERLAY, shape_overlay_present(), shape_dropped_containers() (reads the override file — stdlib line scan, no YAML dep) ; _compose_files() appends `-f docker-compose.shape.yml` LAST (after audio, so its !reset lands last); fleet_containers() excludes the dropped gear. Note: docs confirm `!reset` CLEARS a value (list *replacement* is `!override`); clearing the gateway depends_on is the correct fix here. Requires Docker Compose v2.24+ (compose-spec merge tags). 21 new tests; full suite green. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012Cg9tcyAAC5CmDXuQXDvTU

…reclaim, dropped-lobe honesty (#113) (0.42.0) (#117)

…ntract matrix, live evidence + docs (#112) (0.43.0) (#121) * feat(t3): honest referral — opt-in peer config annotates capabilities + role_infeasible 404s Mesh-brain t3 (issue #112, the confirmed cross-box decision: direct + referral). A box that dropped a role can now name the peer that hosts it: - Peer-config surface: one OPERATOR-DECLARED env var per core role's backend — PRIMARY/MULTIMODAL/EMBED/RERANK_PEER_ORIGIN (PEER_ORIGIN_ENV in lobes/gateway/_config.py), mirroring the *_FEASIBLE idiom. Full origins only, never derived from hostnames/interfaces (the #92 lesson). - With peers declared, GET /capabilities / lobes capabilities (gateway and offline paths) add hosted_by to each unhosted (feasible:false) role, and the 404 role_infeasible body carries the same referral; /v1/models is untouched. - Zero peer config = byte-identical responses to the pre-referral contract (regression-pinned against the exact pre-change bytes). - NO data-plane proxying: the referral is an annotation for the caller to dial directly — the gateway never opens an outbound connection for an unhosted-role request (test-enforced at the handle_post seam and over a real HTTP loopback with a hard no-outbound guard). Proxy-lobes stays deferred to issue #115. Plumbing: RoutingTable.peer_origins (default empty), the shared lobes.roles.annotate_peer_referrals helper (one implementation for both surfaces), gateway compose-template env passthrough + env.example docs, docs/deployment-shapes.md referral section, lobes explain shapes update, and the regenerated template-defaults golden. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013ukvsit214BtGfiTRbdnQC * Add orin-small reference shape: declared-but-unvalidated small-model corner (t2, issue #112) The Jetson AGX Orin 64GB reference for the mesh-brain end-state: hosts NEITHER heavy generate lobe (cortex and senses both absent), keeps the opt-in `minor` gear (vllm-minor) + embedder + reranker + audio. `minor` is not one of the six first-class Colleague roles (SHAPE_ROLES), so the Shape schema's role vocabulary is extended minimally: a new COLLEAGUE_ROLES constant (the six roles machine-as-brain hosts) is split out from SHAPE_ROLES (COLLEAGUE_ROLES + the opt-in OPT_IN_ROLES = ("minor",)) so the identity-shape invariant stays correct now that SHAPE_ROLES is a strict superset. `minor` carries no Profile knobs of its own (like stt/tts) and maps to the already-shipped vllm-minor compose service — zero new compose files, zero core-role render-path branches. Ships as pure declared data per the #108 rule, mirrored exactly from base.toml's unrecognised-card fallback discipline: renders against the existing base/spark/thor card profiles (no new "orin" Profile), goldens only, no live validation claimed anywhere (deployment-shapes.md, the explain-catalog shapes topic, CLAUDE.md, and the --shape CLI help all mark it declared/unvalidated). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013ukvsit214BtGfiTRbdnQC * test(t4): contract-test matrix for the reference shapes — per-(shape, dropped role) honesty cells, referral-aware 404s, t1 golden-value pins One data-driven enumeration of every (built-in shape, dropped core role) cell, derived from the shipped shape TOMLs' hosts lists so a future shape is covered automatically. Per cell: capabilities flag feasible:false on both the gateway payload and the CLI registry, /v1/models omits the role, every TIER_ROLE-derived alias 404s role_infeasible dialing nothing, and with peer origins declared the 404 + capabilities carry the RIGHT per-role hosted_by referral (orin-small's two drops prove referrals never cross). Includes a real-HTTP loopback per cell with an outbound tripwire, hosted- lane liveness per shape (incl. orin-small's minor gear), the honest unspecified-model 404 on orin-small, and the t1 full-native reclaim values (262144 / util>0.30 on spark-lobe; 131072 / util>0.14 on thor-lobe) pinned as explicit assertions against the golden files so a regen cannot silently lower them. No production code changed; machine-as-brain goldens untouched. Part of the mesh-brain end-state plan (issue #112), task t4. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013ukvsit214BtGfiTRbdnQC * t5: mesh-brain acceptance — orin-small + referral phases in accept-shape.sh, live Thor referral evidence (#112 t5) accept-shape.sh gains the orin-small shape arm (dropped-role checks now loop over multiple dropped roles) and an opt-in phase 4b that verifies hosted_by referrals on both honesty surfaces whenever <PREFIX>_PEER_ORIGIN is declared. docs/evidence/2026-07-14-accept-referral-thor.txt: live on the physical Thor — referral 404s + capabilities hosted_by (real declared Spark origin), hosted lanes serving through the same gateway, cross-box cortex reachability, and byte-for-byte shape restore. Full-shape boots reuse the #113 transcripts per operator decision. * docs(t6): mesh-brain end-state — the four #112 decisions, tax table with ratios, live referral evidence (#112 t6) Writes the mesh-brain end-state section of docs/deployment-shapes.md, records all four #112 decisions (direct+referral, cheap-gear co-residence, the Spark/Thor/Orin reference shape assignment, and the mixable shape axis) quoting only measured numbers from the three committed acceptance transcripts, and aligns CLAUDE.md and lobes/explain/catalog.py with the same framing. Fixes a couple of stale cross-references in docs/machine-profiles.md and docs/colleague-stack.md (the "always-on duo" / "four co-resident backends" claims predate the shape axis) and the accept-shape.sh usage block (missing orin-small). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013ukvsit214BtGfiTRbdnQC * chore: bump version to 0.43.0 + changelog for the #112 mesh-brain end-state implementation Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013ukvsit214BtGfiTRbdnQC * evidence: thor-lobe applied for keeps on the physical Thor — full unattended PASS incl. referral phase (#112 t5) Live PR test: this box now serves Gemma senses (131072 @ util 0.30) and not Qwen — cortex feasible:false with hosted_by http://spark.tail0be7e0.ts.net:8001 on both surfaces and in the 404 body. Gateway runs the 0.43.0.dev239 TestPyPI build of this branch. The re-scaffold also picked up the #110 reranker knobs: rerank_relevance PASS (ordering [0,2,1], 1.45s) — resolving the stale-deployment hang recorded in 2026-07-14-accept-referral-thor.txt and #105/#119. First unattended attempt hit the documented unified-memory boot race (21.89 GiB free vs 36.85 needed; 16 restarts) — drop_caches freed 96 GiB and this second run passed end-to-end (fleet healthy in 282s). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013ukvsit214BtGfiTRbdnQC * eidetic: record the Thor's permanent move to thor-lobe (Gemma-only, cortex referred to the Spark) (#112/#121) * chore: uv.lock 0.43.0 + agex pr events * fix: orin-small actually activates its minor lane — render COMPOSE_PROFILES + gateway wiring (Qodo find, #112 t2) Hosting an opt-in role now renders its activation env in shape_env (OPT_IN_ACTIVATION_ENV mirroring the template defaults, kept honest by a mirror test): COMPOSE_PROFILES=minor un-gates the profile-gated vllm-minor service and MINOR_BASE_URL/MINOR_SERVED_NAME wire the gateway backend. Verified end-to-end: a scaffolded orin-small's compose config now includes vllm-minor (was: gateway+pooling only, no generate lane). orin-small goldens +3 lines each; machine-as-brain/spark-lobe/thor-lobe goldens byte-identical. Also: accept-shape.sh validates its non-core CLI deps up front (curl, python3, docker, uv) per the review's skill insight. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013ukvsit214BtGfiTRbdnQC --------- Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

OriNachum and others added 2 commits July 14, 2026 07:24

OriNachum and others added 2 commits July 14, 2026 07:50

Bump 0.41.0 -> 0.41.1 (re-apply brain-shapes spec+plan changelog on m…

4d4fbff

…erged base) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01U9ybFURhQxT9UpCgQNVL67

OriNachum temporarily deployed to testpypi July 14, 2026 04:52 — with GitHub Actions Inactive

OriNachum temporarily deployed to testpypi July 14, 2026 05:12 — with GitHub Actions Inactive

OriNachum changed the title ~~Deployment shapes: machine-as-brain vs mesh-brain lobe profiles — spec + plan (0.40.4)~~ Deployment shapes: machine-as-brain vs mesh-brain lobe profiles — spec + plan (0.41.1) Jul 14, 2026

OriNachum merged commit bad6bb0 into main Jul 14, 2026
9 checks passed

OriNachum deleted the spec/brain-shapes branch July 14, 2026 05:17

This was referenced Jul 14, 2026

Audio lanes during warm-up: 502 without Retry-After, ready-flag disagreement, and feasible:true on no-audio scaffolds #118

Open

Pressure policy sheds 100% of generate traffic forever: swap OCCUPANCY is sticky, PSI shows zero stall #100

Open

OriNachum added a commit that referenced this pull request Jul 14, 2026

Deployment shapes implementation: Shape schema, --shape init, budget …

8aa2d2f

…reclaim, dropped-lobe honesty (#113) (0.42.0) (#117)

OriNachum mentioned this pull request Jul 14, 2026

Mesh-brain end-state: orin-small shape, honest cross-box referral, contract matrix, live evidence + docs (#112) (0.43.0) #121

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Deployment shapes: machine-as-brain vs mesh-brain lobe profiles — spec + plan (0.41.1)#113

Deployment shapes: machine-as-brain vs mesh-brain lobe profiles — spec + plan (0.41.1)#113
OriNachum merged 5 commits into
mainfrom
spec/brain-shapes

OriNachum commented Jul 14, 2026 •

edited

Loading

Uh oh!

OriNachum commented Jul 14, 2026

Uh oh!

qodo-code-review Bot commented Jul 14, 2026 •

edited

Loading

Uh oh!

qodo-code-review Bot commented Jul 14, 2026

Uh oh!

qodo-code-review Bot commented Jul 14, 2026

Uh oh!

sonarqubecloud Bot commented Jul 14, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Conversation

OriNachum commented Jul 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

The two shapes

Aligned to the shipped #110 substrate (updated mid-flight)

Evidence this stands on

The plan: 8 tasks, 5 waves

Uh oh!

OriNachum commented Jul 14, 2026

Uh oh!

qodo-code-review Bot commented Jul 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Code Review by Qodo

New Review Started

Uh oh!

qodo-code-review Bot commented Jul 14, 2026

PR Summary by Qodo

Uh oh!

qodo-code-review Bot commented Jul 14, 2026

Code Review by Qodo

Great, no issues found!

Uh oh!

sonarqubecloud Bot commented Jul 14, 2026

Quality Gate passed

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

OriNachum commented Jul 14, 2026 •

edited

Loading

qodo-code-review Bot commented Jul 14, 2026 •

edited

Loading