docs: close the v1.0 parity gaps (#1081, #1083, #1084, #985)

docs/getting-started/baremetal-ubuntu.mdx

@@ -64,9 +64,13 @@ targets a 128 GB box and leaves only ~4 GB for the host — scale it down to
 reserve more host RAM. After rebooting, confirm with `cat /proc/cmdline`.
 
 <Aside type="caution">
-The old `amdgpu.gttsize` parameter is **deprecated** — it is ignored on
-Linux 6.10+ (the GTT window is now driven entirely by `ttm.pages_limit`).
-Set only `ttm.pages_limit`; drop `amdgpu.gttsize` from any older guide.
+The old `amdgpu.gttsize` parameter is **deprecated** and may emit kernel
+warnings on current amdgpu/TTM kernels — the GTT window is now driven
+solely by `ttm.pages_limit`. Some newer kernels expose the same knob as
+`amdttm.pages_limit` (the out-of-tree AMD TTM module) instead — check
+which one your kernel ships with `modinfo ttm` / `modinfo amdttm` and set
+whichever applies. Set only that knob; drop `amdgpu.gttsize` from any
+older guide.
 </Aside>
 
 ## 3. Install hal0

docs/getting-started/first-model.mdx

@@ -25,8 +25,14 @@ shows up in a few places:
 - The **Models** view has an empty registry (the Inference tab reads "the
   catalog is empty"), and the Downloads panel prompts you to add a model
   from the catalog or via **Add by HF coords**.
-- The **Slots** view shows a **no slots configured** hint if no capability
-  slots exist yet.
+- **OpenWebUI** (`:3001`) is reachable, but its model selector is empty — it
+  reads the same empty `/v1/models` list below.
+- The **Slots** view, `hal0 status`, and `hal0 slot list` show the
+  capability slots the installer scaffolded (structure only — device,
+  profile, port) with no model assigned, rather than an empty list: the
+  installer creates slot *shape*, never model picks. The one case where
+  these are truly empty is an install that skipped slot scaffolding
+  entirely with `hal0 setup --no-slots`.
 - `GET /v1/models` returns an empty list, and a chat request against an
   unconfigured slot has nothing to dispatch to.
 
@@ -51,6 +57,17 @@ hal0 model run  qwen3-4b      # assign to a slot, load, wait until serving
 polls until the server answers, and prints a copy-paste `curl` smoke test
 against the slot port. That's a working chat endpoint.
 
+<Aside type="tip" title="Where the pull lands">
+Downloads write to `[models].store` in `hal0.toml` — the single directory
+the pull engine writes to and slot containers bind-mount. It's empty by
+default and falls back to the legacy `pull_root` (`/var/lib/hal0/models`)
+until you set it explicitly with `hal0 setup --storage-dir <path>`
+(first-run setup) or `hal0 model store <path>` (any time — see
+[Already have the files on disk?](#already-have-the-files-on-disk) below).
+See [Paths and files](/docs/reference/paths-and-files/) for the full
+resolution order.
+</Aside>
+
 ## Already have the files on disk?
 
 Many installs start from a disk full of GGUFs, not a download. Point hal0

docs/getting-started/proxmox-lxc.mdx

@@ -96,9 +96,13 @@ other tenants. Scale it down to reserve more host RAM. After rebooting,
 </Aside>
 
 <Aside type="caution">
-The old `amdgpu.gttsize` parameter is **deprecated** — it is ignored on
-Linux 6.10+ (the GTT window is now driven entirely by `ttm.pages_limit`).
-Set only `ttm.pages_limit`; drop `amdgpu.gttsize` from any older guide.
+The old `amdgpu.gttsize` parameter is **deprecated** and may emit kernel
+warnings on current amdgpu/TTM kernels — the GTT window is now driven
+solely by `ttm.pages_limit`. Some newer kernels expose the same knob as
+`amdttm.pages_limit` (the out-of-tree AMD TTM module) instead — check
+which one your kernel ships with `modinfo ttm` / `modinfo amdttm` and set
+whichever applies. Set only that knob; drop `amdgpu.gttsize` from any
+older guide.
 </Aside>
 
 ## 2. Create the container

docs/guides/configure.mdx

@@ -120,8 +120,8 @@ Advanced, About.
   `[memory]`, and `[activity]` key that previously required `hal0 config
   edit` (including `slots.evict_pressure_mb`, the pressure-eviction floor;
   the `memory.engine` picker only offers `hindsight`/`pgvector` even though
-  the schema also accepts `cognee`/`mem0`, since those two are silently
-  remapped to hindsight), plus a dedicated **memory graph** panel for
+  the schema also accepts `mem0`, since it's silently remapped to
+  hindsight), plus a dedicated **memory graph** panel for
   `[memory.graph]` (`extraction_slot`, `llm_timeout_s`) and a **Restart
   hal0-api** button — a one-click restart-and-health-poll for
   `hal0-api.service`, handy after any change flagged "restart required".

docs/guides/enable-memory.mdx

@@ -1,6 +1,6 @@
 ---
 title: Enable memory
-description: hal0's memory subsystem is gated by [memory].enabled in hal0.toml (on by default). Toggle it with hal0 memory enable/disable, control graph extraction with hal0 memory graph --slot, and migrate any legacy store.
+description: hal0's memory subsystem is gated by [memory].enabled in hal0.toml (on by default). Toggle it with hal0 memory enable/disable and control graph extraction with hal0 memory graph --slot.
 sidebar:
   order: 50
 ---
@@ -132,23 +132,6 @@ unaffected either way — this gate only controls the graph layer.
 </TabItem>
 </Tabs>
 
-## Migrate a legacy store
-
-If you previously ran the Cognee-backed memory store, preview a migration
-into Hindsight banks:
-
-```sh
-hal0 memory migrate --dry-run
-hal0 memory migrate --dry-run --cognee-dir /var/lib/hal0/memory/cognee
-```
-
-The report shows rows total / mapped / unmapped and whether it's a no-op.
-
-<Aside type="caution">
-Migration is **dry-run only** today — it reports the plan without writing.
-Passing `--no-dry-run` exits with a "not yet implemented" error.
-</Aside>
-
 ## Destructive ops are audited
 
 Every destructive call under `/api/memory/*` — bank delete, and the
@@ -167,3 +150,6 @@ confirmation is rejected with `memory.confirm_required`.
   the memory store.
 - [MCP tools](/docs/reference/mcp-tools) — the `memory_*` tool tiers and
   the `/mcp/memory` server this guide's flag gates.
+- [Honcho memory](/docs/guides/honcho-memory) — `hal0 memory migrate --from
+  <engine> --to <engine>` moves an agent's memories between Hindsight and
+  Honcho.

docs/reference/config-schema.mdx

@@ -116,14 +116,16 @@ NPU weights off the root filesystem too.
 
 | Key | Type | Default | Notes |
 |---|---|---|---|
-| `engine` | str | `"hindsight"` | Active memory engine: `hindsight` (default) \| `mem0` \| `pgvector`. `cognee` is **deprecated** — accepted for back-compat but resolves to `hindsight`. |
+| `engine` | str | `"hindsight"` | Active memory engine: `hindsight` (default) \| `mem0` \| `pgvector`. `cognee` was **retired in v1.0.0** — the schema now rejects it (the Cognee store has been dark since v0.4). |
 
 <Aside type="note">
-The validator also accepts `cognee`/`mem0`, but the Advanced Settings dropdown
-only offers `hindsight`/`pgvector` — the factory silently remaps `cognee`/`mem0`
-selections to `hindsight`, so the dashboard doesn't offer a choice that would
-lie about what actually runs. Hand-editing `hal0.toml` to `cognee`/`mem0` is
-still accepted at the schema level, but resolves to `hindsight` in practice.
+The validator accepts `mem0`, but the Advanced Settings dropdown only offers
+`hindsight`/`pgvector` — the factory silently remaps `mem0` selections to
+`hindsight`, so the dashboard doesn't offer a choice that would lie about
+what actually runs. Hand-editing `hal0.toml` to `mem0` is still accepted at
+the schema level, but resolves to `hindsight` in practice. `cognee` is no
+longer a valid value at all — a `hal0.toml` still carrying it fails config
+validation and must be updated to `hindsight`.
 </Aside>
 
 ### `[memory.embedding]` — reranker knobs (Hindsight era)

installer/agents/hermes/plugins/hal0-provider/profile.py

@@ -8,8 +8,9 @@
   purely *declarative*: it advertises the endpoint, capabilities, and a live
   model inventory; it does not itself stream tokens.
 * ``base_url = "http://127.0.0.1:8080/v1"`` — loopback hal0-api (never a slot
-  port). No Bearer on the hal0 LAN (post-ADR-0012 identity model); identity is
-  carried by ``X-hal0-Agent`` when we make our own discovery call.
+  port). No Bearer on the hal0 LAN — hal0 ships with no built-in auth on the
+  trusted local network; identity is carried by ``X-hal0-Agent`` when we make
+  our own discovery call.
 * ``default_aux_model = "hal0/agent"`` — pins compression / vision /
   summarization / web_extract aux calls at the local agent slot. ``hal0/agent``
   is a *virtual* name hal0 resolves per-request via its LiveSlotResolver, so

src/hal0/agents/hermes/plugins/provider_hal0/profile.py

@@ -8,8 +8,9 @@
   purely *declarative*: it advertises the endpoint, capabilities, and a live
   model inventory; it does not itself stream tokens.
 * ``base_url = "http://127.0.0.1:8080/v1"`` — loopback hal0-api (never a slot
-  port). No Bearer on the hal0 LAN (post-ADR-0012 identity model); identity is
-  carried by ``X-hal0-Agent`` when we make our own discovery call.
+  port). No Bearer on the hal0 LAN — hal0 ships with no built-in auth on the
+  trusted local network; identity is carried by ``X-hal0-Agent`` when we make
+  our own discovery call.
 * ``default_aux_model = "hal0/agent"`` — pins compression / vision /
   summarization / web_extract aux calls at the local agent slot. ``hal0/agent``
   is a *virtual* name hal0 resolves per-request via its LiveSlotResolver, so

src/hal0/agents/manager.py

@@ -315,9 +315,9 @@ def install(
             # Already installed — return the existing record. Idempotent.
             return self._read_record(name)
 
-        # Single-pick (ADR-0004 §2): only one bundled agent may be installed
-        # at a time. ``blocking`` == ``current`` — anything already
-        # installed conflicts with a fresh install.
+        # Single-pick: only one bundled agent may be installed at a time.
+        # ``blocking`` == ``current`` — anything already installed conflicts
+        # with a fresh install.
         blocking = current
 
         if not blocking:

src/hal0/config/schema.py

@@ -2446,16 +2446,22 @@ class MemoryConfig(BaseModel):
             "(private:<agent> / project:<id> banks + fan-out recall)."
         ),
     )
-    # HAL0-SUNSET: v1.0.0 — 'cognee' engine value resolves to hindsight at runtime; drop the alias.
+    # HAL0-SUNSET: v1.0.0 — 'cognee' engine literal retired here. It is no
+    # longer an accepted config value (was previously a back-compat alias
+    # that resolved to hindsight at runtime). provider_from_config's runtime
+    # resolution still treats any *unrecognized* engine as Hindsight as a
+    # defense-in-depth fallback for callers that bypass this validator
+    # (e.g. duck-typed namespaces), but a hal0.toml with engine = "cognee"
+    # now fails config validation outright — the Cognee store has been dark
+    # since v0.4.
     engine: str = Field(
         default="hindsight",
         description=(
             "Active memory engine. 'hindsight' (default) is the platform "
-            "engine; 'mem0' and 'pgvector' are alternates. 'cognee' is "
-            "DEPRECATED — the legacy Cognee store has been dark since v0.4 and "
-            "its wrapper was removed; the value is still accepted for "
-            "back-compat but resolves to 'hindsight' at runtime. Use "
-            "'hindsight'."
+            "engine; 'mem0' and 'pgvector' are alternates. The legacy "
+            "'cognee' value was retired in v1.0.0 (the Cognee store has been "
+            "dark since v0.4 and its wrapper was removed) — a hal0.toml "
+            "still carrying it must be updated to 'hindsight'."
         ),
     )
     agent_providers: dict[str, str] = Field(
@@ -2482,8 +2488,8 @@ class MemoryConfig(BaseModel):
     @field_validator("engine")
     @classmethod
     def _engine_is_known(cls, v: str) -> str:
-        known = {"cognee", "hindsight", "mem0", "pgvector"}
-        s = str(v or "cognee").strip().lower()
+        known = {"hindsight", "mem0", "pgvector"}
+        s = str(v or "hindsight").strip().lower()
         if s not in known:
             raise ValueError(f"memory.engine {v!r} must be one of {sorted(known)}")
         return s

tests/config/test_memory_engine_field.py

@@ -13,10 +13,18 @@ def test_engine_defaults_to_hindsight():
 
 
 def test_engine_accepts_known_engines():
-    for e in ("cognee", "hindsight", "mem0", "pgvector"):
+    for e in ("hindsight", "mem0", "pgvector"):
         assert MemoryConfig(engine=e).engine == e
 
 
 def test_engine_rejects_unknown():
     with pytest.raises(ValidationError):
         MemoryConfig(engine="weaviate")
+
+
+def test_engine_rejects_retired_cognee_literal():
+    # HAL0-SUNSET: v1.0.0 — 'cognee' was a back-compat alias that resolved to
+    # hindsight at runtime; it is now retired and fails validation like any
+    # other unknown engine. The Cognee store has been dark since v0.4.
+    with pytest.raises(ValidationError):
+        MemoryConfig(engine="cognee")

ui/src/dash/memory-map.jsx

@@ -121,7 +121,7 @@ export function useMemoryMapModel() {
   const slots = slotsQ.data || []
 
   // Pool total = the real ceiling for GPU model loads. On UMA (Strix Halo)
-  // that is the GTT cap (amdgpu.gttsize, ~80 GiB — live as
+  // that is the GTT cap (ttm.pages_limit, ~80 GiB — live as
   // stats.gpu_vram_total_mb), NOT the full unified RAM (128 GiB): models can't
   // actually allocate past the GTT window, so headroom must be measured
   // against it. Prefer the live GTT cap; fall back to the static unified/ram