The host now depends on openhuman-rpc alone instead of naming both
openhuman-rpc and openhuman_core directly, with the product gates forwarded
on that single dependency so the embedded core, session owner, and transport
all arrive through the crate chain. The e2e-test-support gate is likewise
forwarded to openhuman-rpc, and the feature comments were consolidated to
document why each gate must stay listed.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

Adds an e2e-test-support feature to the embed, rpc, and tinyhumans crates that forwards to the core crate, letting the desktop E2E build reach the destructive openhuman.test_reset RPC through the dependency chain instead of naming core directly. The feature stays default-off everywhere.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

The CLI binary now depends only on openhuman-rpc, with core, embed and
tinyhumans moved to dev-dependencies so they build into tests and examples
but never into the shipped binary. Feature gates forward to openhuman-rpc
alone, and the benchmark bins that named engine crates directly were
dropped.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

Introduce a library_profile binary that drives scripted agent scenarios
through a mock runtime to measure library-level behaviour, along with
rss, tool dialect, and tool search benchmark binaries for tracking
resource usage and tool selection quality.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

…pty, and the stat is blank to

Please paste the diff (or at least the stat and a summary of what changed) and I'll write the Conventional Commits message.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

Adds unit tests covering the CLI main entry point to guard against regressions in argument handling and startup behaviour.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

The embed examples now build through `openhuman_embed::Runtime` and dispatch
RPC via `Runtime::core_runtime()` instead of the `CoreBuilder`/`CoreRuntime`
pair, matching the library surface embedders are meant to use. The headless
example's docs also drop the stale `serve` guidance in favour of the
`openhuman-rpc` host entries, and both examples note that the default
workspace is ephemeral.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

The TUI now declares its shipped feature set explicitly and forwards every
gate to openhuman-rpc instead of pulling in the core's whole default set
implicitly. Direct dependencies on openhuman-core, openhuman-tinyhumans,
dotenvy and sentry were dropped since those facades are reached through
openhuman-rpc.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

The TUI now builds its Sentry client from embed's shared option chain
instead of duplicating the noise filters, secret scrubbing, PII defaults
and transport, so all three binaries report crashes identically. Only the
DSN, release tag and environment are still resolved locally, and
resolve_environment now takes the app env as a parameter so it can be
tested without touching process state.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

Introduce a session module and state handling so the TUI can track
conversation state across interactions. Add controls and rendering
support to drive the new session flow.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

Expose the runner module from the crate root so its functionality is
reachable to consumers of the library.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

Removed the `openhuman-core` workspace dependency entry, which was
unintentionally pulling the core crate's full default feature set into any
member that named it. Added a comment documenting the library chain and
noting that hosts should depend on `openhuman-rpc` only.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

Removed the dhat, mintex, thousands, and rustc-hash 1.1.0 entries along with the dotenvy, sentry, tinyjevclient, and tinytools-jev dependencies, and added openhuman-embed to the main crate. The lockfile now resolves rustc-hash to a single 2.1.3 version.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

Cleaned up the app crate by dropping imports and helper code that are no longer referenced, reducing compiler warnings without changing behaviour.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

Introduce a local data reset flow and workspace path resolution in the
app crate, along with file logging and session command updates. These
additions support clearing local state and locating workspace files
consistently across the desktop app.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

Moved the process spawning and stdio wiring out of the main start
routine into a dedicated helper so the startup path reads more clearly
and the spawn behaviour can be reused. No behaviour change.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

Introduce a module to spawn and supervise the core process, tracking its
state and handling shutdown so the app can manage the backend lifecycle
reliably.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

Introduce a core process module that spawns and supervises the backend
process, tracking its lifecycle and exposing status to the app. Tests
cover startup, shutdown, and failure handling.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

The comment describing the in-process core now points at the current RPC host entry point, and the lockfile reflects the dependency set that entry point pulls in.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

…hain

The Sentry noise filter now lives in embed's shared before_send chain, so
these tests assert the shell's rule is still carried under its
localhost-dev-fetch name. A helper resolves the event through the shared
known_noise function to keep the existing assertions valid.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

The shell now declares its openhuman dependency as `openhuman-rpc` rather than
`openhuman_core`, so the product-forwarding guard parses that dependency and
exempts rpc's own local gates (`http-client`, `server`, `jev`) from the
unexpected set. The library chain is updated to rpc → cli/tui, and a new
`rpcForwardedGates` helper exposes the gates a host must forward.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

The forwarding check now validates the tui crate alongside cli, and both hosts
are asserted against rpc's forwarded gates rather than core and tinyhumans
directly, matching the hosts' current dependency on rpc alone.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

The header comments now describe the shell's dependency on openhuman-rpc
and the cli and tui hosts on rpc, and the usage line lists the tui
manifest argument. This keeps the documentation in step with the
manifests the check actually inspects.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

Rework the forwarding tests around the openhuman-rpc hop, asserting that rpc's own local gates are not mistaken for product gates and that hosts forwarding straight to core are reported as misrouted. The checked-in manifest test now also walks openhuman-tui so the whole chain is verified.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

Add a script that verifies the crate dependency chain in CI, catching
violations of the intended layering before they reach the main branch.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

Add tests for the crate chain checker covering missing manifests, broken
dependency links, and valid chains so regressions in the validation logic
are caught early.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

The crate chain test no longer imports cpSync and drops the assertion that
checked it was a function, since the import was only kept around for an
older lint configuration.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

The rust:layout script now also runs the crate chain check after the layout check, so both validations run together under the existing command.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

The scripts/profile directory and its six driver scripts are gone, along with the README that documented them. These benchmarks were run manually rather than in CI, and the library-profile scenarios they wrapped no longer exist in the binary.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

The README now lists check-crate-chain.mjs alongside check-openhuman-rust-layout.mjs as scripts wired to a pnpm script, so the local running instructions match the current set of wired checks.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

The crate-chain checker now runs as an always-on static gate in the
self-hosted lane plan, since a manifest edit anywhere can break the
core -> embed -> tinyhumans -> rpc -> app/cli/tui chain. It is also
registered in the rust-core paths filter so changes to the checker
trigger the relevant lanes.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

The crate layout section now spells out the core → embed → tinyhumans → rpc → hosts chain, with each crate depending only on the layer below and hosts reaching the core through the rpc facade. Host boot entry points, feature-gate forwarding, and the TinyHumans install path were updated to match, so contributors follow the enforced layout instead of the old direct dependencies.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

Reword the desktop core and CLI server notes so the launcher path is
described in terms of `openhuman_rpc::host::cli`, with the older
`install_cli_server()` and `run_core_from_args` pair noted as the
equivalent for embedders that predate it.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

The crate overview now describes the layering as a strict chain where each
crate depends only on the layer directly below it, with hosts booting through
the shared `openhuman_rpc::host` entry points and reaching core internals via
embed's doc-hidden `__host` list. The dependency table, layering diagram and
feature-forwarding notes were updated to match, and the new
`check-crate-chain.mjs` guard is called out so contributors know which edges

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

The architecture pages now describe the strict dependency chain from the
three hosts down through openhuman-rpc, openhuman-tinyhumans and
openhuman-embed to openhuman-core, and note that hosts boot through
openhuman_rpc::host and reach the core only via the curated facade. The
Tauri shell page gains a table mapping each shell concern to the rpc
entry point it uses, plus the CI checks that enforce the chain.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

The crate README now describes `openhuman_rpc::embed` alongside `tinyhumans` as the curated host facade and notes that hosts forward their feature gates here 1:1. Consumer notes were refreshed to reflect the `host::cli`, `host::desktop` and `host::tui` entry points, and the session-store README drops the stale claim that the TUI still calls `install()` directly.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

…ed facades

The crate now depends on openhuman-rpc instead of openhuman-core directly, so
the boot sequence, runtime, Sentry, logging and bundled-modules sections were
rewritten to name the host and embed entry points that forward to the core.
The dependency and feature-gate notes were also updated to reflect the single
forwarding dependency and the rpc-owned http-client, server and jev features.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

The startup walkthrough, feature table and boundaries section now describe
the core booting through `openhuman_rpc::host::tui()` and the `embed`
facades instead of direct `openhuman-core` calls, and the feature table
lists the contributor gates forwarded to `openhuman-rpc`. This keeps the
documentation aligned with the crate's new dependency chain.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

The README now points hosts at `embed::process::tokio_runtime()` instead of
building a tokio runtime by hand from core constants, so no core dependency is
needed to name the stack size and blocking thread limits. The examples section
also reflects that the headless and kernel examples drive this crate's
`Runtime` rather than `CoreBuilder` directly.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

Reformatted the port-conflict error match in the core process handle and the
release tag construction in crash reporting so both conform to rustfmt's line
width. No behaviour changes.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

Update the vendored submodule pointers for tinyagents, tinymcp, tinysearch and tinyskills to their latest upstream commits.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

The runtime builder now checks whether the discovered config was unavailable before installing a live policy, since the placeholder base config points at the default root rather than the operator's install and would guard the wrong tree. In that case it logs a warning and keeps the policy installed by the core's bootstrap.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

Update the pinned submodule commits for the four vendored tiny libraries to
their latest revisions.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>

…ease

The shutdown path referenced the orchestration module through the
openhuman_core crate path, which does not resolve from within this crate.
Switching to the crate-relative path lets the background completion stores
be released so a data reset can delete the workspace directory.

Auto-committed-on: dragonfly
Co-authored-by: Medulla <medulla@tinyhumans.ai>