The supermemory SDK already honors SUPERMEMORY_BASE_URL, but the raw
urllib call used for session-end conversation ingest hardcoded
https://api.supermemory.ai/v4/conversations, so ingest always hit the
cloud even when pointing at a self-hosted server (e.g.
http://localhost:6767).
Resolve the base URL as config (supermemory.json base_url) >
SUPERMEMORY_BASE_URL env var > https://api.supermemory.ai, strip any
trailing slash, and use it for both the SDK client and the
/v4/conversations ingest endpoint.
The staleness check's bespoke mtime memo keyed only on the user
config.yaml, but load_config() merges the managed-scope config
(HERMES_MANAGED_DIR/config.yaml, /etc/hermes) whose leaf keys win. A
managed honcho.timeout with no user config.yaml made the memo cache
'no timeout' while _build resolved the managed value — the same
perpetual-rebuild mismatch this PR fixes for honcho.json. A managed
timeout edit was likewise invisible while the user file's mtime stayed
put.
load_config_readonly() is already cached on both files' signatures plus
the env-ref snapshot, so use it instead of duplicating that
invalidation logic; the defensive deepcopy the old memo existed to
avoid is skipped by the readonly variant. Drive the rebuild test
through a real config.yaml and add a HERMES_MANAGED_DIR regression
test covering stable reuse and managed-timeout edits.
The staleness check added in #66052 resolved the timeout from env,
config.yaml, and the default only, while the build path also reads the
honcho.json host block (timeout/requestTimeout). With a timeout
configured in honcho.json, the two permanently disagreed: every
no-config get_honcho_client() call — i.e. every HonchoSessionManager
.honcho property access — interpreted the mismatch as a config change
and tore down and rebuilt the client, defeating the singleton on the
hot path it was meant to protect.
Teach the check to read honcho.json through the same host-aware chain
as from_global_config, memoized on the file's mtime_ns so the per-call
cost stays one stat(). A genuine honcho.json timeout change is now also
detected, extending #57437 to that config surface.
Follow-ups for the consolidated salvage:
- Memoize the config.yaml-derived timeout on the file's mtime_ns so the
rebuild-on-timeout-change check from PR #57437 costs one stat() per
get_honcho_client() call instead of a full YAML load on the hot path.
- hermes honcho status now displays the host-block-resolved
dialecticCadence (remnant from PR #63776, whose runtime fix landed
in #62290).
- AUTHOR_MAP entries for the salvaged contributor emails.
The Honcho client singleton cached the HTTP timeout at first build.
In long-lived processes (gateway, dashboard), changing the timeout
via config.yaml or HONCHO_TIMEOUT had no effect until restart.
Track the resolved timeout alongside the cached client and compare
on each get_honcho_client() call. When the timeout differs, reset
the singleton so the next call rebuilds with the new value.
Fixes#57347
Fixes Honcho client initialization for setup-generated configs that store
connection details in a named host block (e.g. "local"). Previously:
- base_url was only read from flat config root, not from host_block.
- resolve_active_host() ignored defaultHost and always used the Hermes
profile key ("hermes"), so the host block lookup returned {} and the
api_key was also lost.
FixesNousResearch/hermes-agent#61661
honcho_reasoning's minimal tier hard-caps Honcho's dialectic output at 250
tokens combined with the model's own hidden reasoning tokens. Confirmed via
direct honcho-api server logs that a multi-fact query ("summarize known
facts about this peer and communication preferences") run at
reasoning_level=minimal gets cut off mid chain-of-thought at exactly
output_tokens=250, before the model ever reaches a synthesized answer.
low/medium/high/max fall back to Honcho's much larger global dialectic
default and don't hit this cap. Since dialecticDynamic is on by default and
the calling model picks reasoning_level itself via this tool parameter, the
fix is to make the tradeoff explicit in the parameter description so the
model defaults to low unless the query is genuinely a single-fact lookup.
Schema-description-only change; the shared dialectic_max_chars truncation
cap in session.py's dialectic_query() is a separate fix (its own PR).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
_fetch_session_context was using user_peer_id directly as observer
when calling _fetch_peer_context, bypassing _resolve_observer_target.
This caused honcho_context to return empty data because the observer
perspective was wrong (user instead of hermes/assistant).
Fixed by resolving observer via _resolve_observer_target(session, 'user'),
consistent with all other call sites (get_peer_card, honcho_search, etc.).
queryRewrite (default off) gates the latest-message rewrite so the
extra auxiliary LLM call is opt-in. firstTurnBaseWait and
firstTurnDialecticWait expose the turn-1 bounded waits in seconds
(0 disables). All three resolve host-block-first like every other
field. Also pins per-host timeout resolution with tests.
Move query_rewrite from the honcho plugin to plugins/memory/ and
rename the auxiliary task key honcho_query_rewrite ->
memory_query_rewrite so any memory provider can use the same
rewrite path and model/timeout config block. No behavior change.
dialecticMaxChars (default 600) is documented as the budget for the dialectic
supplement auto-injected into the system prompt every turn — a small recurring
cost that is correct to bound tightly. But dialectic_query() applied that cap
unconditionally, so explicit honcho_reasoning tool calls — where the model
deliberately spends a turn asking for a synthesized answer — were silently
truncated mid-word to 600 chars with a trailing " …", no error surfaced. The
full answer is returned by Honcho server-side; the clip happens client-side.
The auto-injection path already has its own token-based budget (contextTokens,
enforced in prefetch() via _truncate_to_budget), so the char cap's real job is a
cheap always-on guardrail for that recurring injection. Explicit tool results
are already bounded server-side by Honcho's dialectic MAX_OUTPUT_TOKENS and don't
need the injection cap — sibling tools (honcho_search, honcho_context) don't
post-clip their results either.
Add apply_injection_cap (default True, preserving current behavior) to
dialectic_query(); the honcho_reasoning tool handler passes False so it returns
Honcho's full synthesized answer. Auto-injection is unchanged. Tests cover both
the capped injection path and the uncapped tool path.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
honcho_conclude's delete action was unreachable in practice: no tool ever
surfaced a real conclusion id for the model to pass as delete_id.
honcho_search only searches the separate Message resource space, and the
SDK's ConclusionScope.list()/.query() (which do return real Conclusion.id
values) were never wired into any tool.
Adds an optional list mode to honcho_conclude (query to search, omit to
browse recent conclusions), backed by a new
HonchoSessionManager.list_conclusions(). No new tool, no changes to the
create/delete signatures or their conclusions_of() routing.
injectionFrequency='first-turn' returned empty for the entire
prefetch_context() method on turns 2+, which blocked the dialectic
supplement from being consumed and injected. The dialectic has its
own cadence (dialecticCadence) and must continue to fire and inject
independently of the base context layer.
Now first-turn mode gates only Layer 1 (base context: representation
+ card), letting Layer 2 (dialectic supplement) flow through its
normal consumption path on every turn.
Also fixes all remaining tests that passed dialecticCadence via
cfg_extra={'raw': {...}} to use the typed dialectic_cadence field.
injectionFrequency, contextCadence, and dialecticCadence were read
only via raw.get() which checks root-level keys in honcho.json.
Settings placed inside hosts.<name> (the normal per-host location)
were silently ignored, falling back to defaults.
- Add typed dataclass fields: injection_frequency, context_cadence,
dialectic_cadence with host-block-first resolution chains matching
the pattern used by all other config fields.
- Update HonchoMemoryProvider.initialize() to read from typed cfg
fields instead of cfg.raw directly.
- Fix search_context tests that mocked _honcho directly — the
.honcho property getter calls get_honcho_client() which overwrites
the backing field, so use patch.object on the property instead.
- Add injection_frequency and context_cadence config override tests.
HonchoClientConfig timeout/requestTimeout resolution skipped the per-host config
block, silently dropping a host-scoped timeout and falling through to the global
config.yaml value (or the default). Add the host block at the front of the
resolution chain, consistent with every other field (base_url, api_key, etc.).
Symptom: Honcho logs show a dialectic answer was generated, but Hermes never
injects it — intermittently.
Root cause: the dialectic supplement that queue_prefetch() fires at the end of
turn N is stored pending (fired_at=N) for consumption by turn N+1's prefetch().
But prefetch()'s trivial-prompt guard returned early BEFORE the consumption
block. So when turn N+1's prompt was trivial ('ok', 'yes', 'continue', a slash
command), the ready result was never consumed, and a few turns later the
stale-discard guard dropped it. Generated by Honcho, never seen by the model.
The dependence on 'is the consuming turn trivial?' is why the loss looked random.
Fix: trivial turns now consume and inject a ready, non-stale pending result while
still spending no new work (no base-context fetch, no new dialectic fire). A
trivial ack shouldn't generate context, but it shouldn't destroy an answer
already computed for that exact turn. Extracted the pop+stale-check into a shared
_consume_pending_dialectic() used by both the trivial path and the normal path so
they age-check identically.
Preserved (covered by new tests): genuinely stale results are still discarded on
trivial turns; trivial turns still fire no new work; a trivial turn with nothing
pending still injects nothing.
Adds regression tests for inject-on-trivial and discard-stale-on-trivial.
A brand-new session injected no Honcho context on the user's first message —
the peer card/representation only showed up from turn 2 onward. The base-context
fetch was fired asynchronously and popped in the same synchronous pass, so it
always lost the race on turn 1 (a background thread can't finish inside one
pass), leaving the first response with zero recalled context.
Fetch the base layer (representation + card + summary) synchronously with a
bounded timeout on turn 1 so the peer card is injected immediately; subsequent
turns still consume the background-refreshed result primed by queue_prefetch().
The wait is bounded by _FIRST_TURN_BASE_TIMEOUT and tightened further by a small
configured request timeout (fail-fast deployments / tests).
Two related first-turn/dialectic reliability fixes ride along:
- first-turn dialectic no longer double-fires: if a prewarm .chat() thread is
already in flight from session init, turn 1 waits briefly for it instead of
firing a second (duplicate) call that also blocked the first response. The
first-turn wait is decoupled from a large host timeout (a 60s host timeout
must not block the first response for 60s) via _FIRST_TURN_DIALECTIC_CAP,
while still honoring a tight configured timeout.
- empty-pass propagation guard in multi-pass dialectic: at depth > 1 each pass
feeds the prior pass's output into the next prompt. If a pass returned empty
(e.g. a reasoning model that spent its whole budget thinking), the next prompt
carried a blank assessment (the "empty spot" seen in Honcho request logs). Now
only non-empty prior results feed dependent passes; if all priors are empty,
re-issue the base prompt instead of referencing nothing.
The five Honcho tool schemas had overlapping/misleading descriptions, making it
hard for the model to pick the right one, plus two concrete correctness bugs:
- honcho_context advertised an 'Optional focus query' parameter that the dispatch
never read. The model could pass query= expecting filtering that never happened.
Remove the dead parameter; honcho_context is an honest no-query snapshot. Focused
retrieval now lives in honcho_search (see prior commit).
- honcho_conclude's peer param was described as 'Peer to query' — wrong; it's the
peer the conclusion is ABOUT. Corrected.
Rewrite all five descriptions to give each tool a distinct mental model and
cross-reference siblings:
profile = read/write the compact card (cheapest, no query, no LLM)
search = find what was actually said, ranked, cross-session (cheap, no LLM)
reasoning= ask a question, get a synthesized answer (the only LLM tool; expensive)
context = a fixed session snapshot (no query, no LLM)
conclude = write a durable fact to the profile
Addresses the honcho_context query param half of #29402 (see PR notes for the
divergence from that issue's proposed wire-through approach).
honcho_search routed through search_context() -> peer.context(search_query=),
which returns the peer's standing representation + card. The search_query arg
does not turn that endpoint into a search, so results were effectively
query-independent: the same representation blob regardless of the query. Factual
lookups ('what medication', 'which value did we pick') returned noise.
Rewire search_context() to call the workspace message-search endpoint
(Honcho.search) with a peer_perspective filter: RRF-ranked (hybrid semantic +
full-text) raw message excerpts spanning every session the peer was a member of,
across all authors, membership-time-scoped. This is the cross-session factual
recall primitive.
peer_perspective is chosen over the alternatives because it is the only scope
that is simultaneously (a) cross-session, (b) inclusive of assistant-authored
facts about the peer (peer-author search drops these, and they are a large part
of what you want to recall about yourself), and (c) privacy-scoped to the peer's
own sessions (plain workspace search leaks other peers' sessions).
- snippets are labeled by author so the model can tell user-stated facts from
assistant-derived ones
- max_tokens is now an enforced budget (was accepted but meaningless)
- graceful fallback to peer-authored search if peer_perspective is unsupported
- query length clamped under the embedding input cap
Replaces 3 change-detector tests that asserted the old representation-dump
behavior with 4 that assert the message-search contract + fallback path.
Conflict resolutions:
- web_server.py: keep the branch's unified provider-config handlers
(declared schema served by default, profile-scoped) over main's
surface=declared query param. Main's declared-surface helpers and the
hermes_cli.memory_providers import are dropped — the branch deleted
that module when provider schemas moved into their plugins, so main's
code path could no longer import.
- hermes.ts: keep profileScoped() calls without ?surface=declared to
match the unified backend.
- constants.ts: take main's new reasoning effort values (max, ultra),
keep the branch's memory provider ordering.
- config-settings.tsx: take main's FallbackModelsField import, drop the
now-unused SECTIONS import (branch replaced it with
sectionFieldEntries).
- provider-config-panel.test.tsx: file moved to settings/memory/ on the
branch; main's act() fixes targeted the old tests, and the rewritten
suite produces no act() warnings, so the old path stays deleted.
- test_web_server.py: keep both sides' new tests (main's MoA endpoint
tests plus the branch's Honcho provider tests).
The dashboard's dedicated memory-provider UI (4b184cbe5) excluded
memory.provider from /api/config/schema server-side. Desktop's settings
page builds its field list from that schema, so the Memory Provider
dropdown silently vanished from Desktop after v0.18.1.
- web_server.py: restore memory.provider as a select in _SCHEMA_OVERRIDES,
with options built from plugins.memory discovery (was a stale hardcoded
[builtin, honcho] list before the removal)
- plugins/memory: add list_memory_provider_names() — directory-scan-only
name listing, safe at module import time (no provider imports)
- web ConfigPage: hide memory.provider client-side instead — the Plugins
page owns the dedicated provider-switching UI there
- tests: schema contract (select present, category memory, builtin
sentinel) + invariant that every discoverable provider is selectable
No bundled provider declares actions, and the motivating request
(openviking, #56309) needs dynamic select options rather than actions.
Removing the unconsumed POST dispatch surface keeps this PR focused on
the config panel; the extension point can return as its own PR with its
first real consumer.
Follow-ups for salvaged PR #43819: the registry key was
str(Path(db_path).expanduser()) — a symlinked or relative path to the
same DB file got its own connection, silently reintroducing the exact
multi-writer contention the registry prevents. Key on Path.resolve()
(OSError-tolerant fallback). Adds a symlink regression test and the
AUTHOR_MAP entry for adambiggs.
Every MemoryStore instance opened its own SQLite connection guarded by
its own RLock. Several providers coexist in one process (the main agent
plus every delegate_task subagent), so instances pointing at the same
memory_store.db raced as independent WAL writers. Combined with writes
that were not rolled back on error, one connection could leave an open
write transaction that pinned the write lock and made every other
connection's writes fail with "database is locked" for the full busy
timeout.
Instances for the same database now share ONE process-wide connection
and ONE re-entrant lock, so access is fully serialized and
cross-connection contention is impossible. The shared connection is
refcounted: closing one instance never tears it out from under a live
sibling, and the last close releases it. The connection runs in
autocommit (isolation_level=None) so a write that raises mid-method can
never leave a dangling transaction holding the write lock; the existing
explicit commit() calls become harmless no-ops.
The provider's shutdown() now calls the refcount-guarded close() instead
of just dropping the reference: leaving finalization to GC kept the
connection (and its write lock) alive indefinitely on long-running
gateways, prolonging the exact contention this fix removes. The last
provider now releases the connection deterministically while siblings
stay live; regression tests fail without the wiring.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Main's #60569 (dashboard memory provider switching) rewrote the same
GET/PUT /api/memory/providers/{name}/config routes this PR owns — the
dashboard and desktop share one backend. Resolve by dispatching: providers
that declare a config_schema.py get the declared path (host-block storage,
locked honcho writes, profile scoping, actions); everything else keeps
main's instance get_config_schema()/save_config path unchanged, so the
dashboard's PluginsPage behavior is preserved for instance providers.
Setup manifests, the /setup endpoint, provider switching, and name
validation from #60569 apply to both paths. Instance payloads gain the
docs_url/actions keys the desktop panel expects; declared payloads gain
the setup block the dashboard expects. Main's hindsight/honcho instance-
schema tests are updated for the dispatch, and its status test gets the
HOME pin it was missing (it read the developer's real ~/.honcho).
The salvaged SelfHostedBackend made self-hosted servers reachable via
mem0.json / MEM0_HOST, but the setup wizard still offered only Platform
and OSS — exactly the gap users hit (Discord report: 'At memory setup
there's only 2 options'). Adds a third wizard mode:
- interactive picker: Platform / Self-hosted server / Open Source
- non-interactive: hermes memory setup mem0 --mode selfhosted
--host http://... [--api-key ...] [--dry-run]
- host -> mem0.json (behavioral), API key -> .env as MEM0_API_KEY
(secret), optional key for AUTH_DISABLED servers
- best-effort reachability check against the server, non-fatal
- README + memory-providers docs updated with the wizard path
Review follow-ups on the salvage:
- get_all() pruned from the ABC and all three backends: mem0_list (its
only caller) was removed by the recall-tuning commit, leaving new,
tested, unreachable code — including SelfHostedBackend's _MAX_TOP_K
over-fetch workaround. Tests for it dropped; fake-class stubs remain
harmlessly. (The #52921 true-total fix lives on in the PR history if
a lister ever returns.)
- The persisted rerank config key was write-only (setup prompted for it,
nothing read it). initialize() now parses it into _rerank_default and
mem0_search uses it when the model doesn't pass rerank explicitly;
per-call args still win. Guard test added.
- Platform-mode setup now warns when MEM0_HOST is set in the environment:
the json host-clear can't help there (_load_config seeds host from the
env var, docs tell users to put it in .env) — the user would silently
keep routing to the self-hosted server.
- SelfHostedBackend: connect-level retries (httpx.HTTPTransport(retries=2))
so a single transient blip doesn't count toward the provider breaker;
transport now injectable and the test helper uses the real __init__
instead of mirroring it via __new__.
- plugin.yaml description no longer leads with reranking (off by default,
platform-only); docs em-dash typo fixed.
Follow-up on the salvaged #55614. The PR added host-based routing to
_create_backend (precedence: oss > host > platform) but two sibling surfaces
didn't mirror it:
- system_prompt_block() checked host before oss, so an oss+host config ran
OSS but told the model it was self-hosted HTTP. Reordered to match routing.
- Platform setup (hermes memory setup mem0 --mode platform) left a stale host
in mem0.json; since host beats platform, the user kept routing to the
self-hosted server. save_config merges (no delete), so clear host to ""
rather than pop() so the merge actually overwrites it.
Adds regression tests for both (mutation-checked).
Salvage of #55614 by @kartik-mem0 (mem0 maintainer). Adds a SelfHostedBackend
that talks to a self-hosted Mem0 Docker server over httpx (X-API-Key auth,
/search + /memories routes), gated behind `host`. Also folds in the mem0
research-team recall tuning that rides with it: rerank defaults to false across
all modes, the mem0_list tool is removed (5->4 tools), search guidance is
de-shouted, and self-hosted get_all reports the true stored total (#52921).
Supersedes the self-hosted portion of #52487 (@liuhao1024, first-submitted).
Closes#52478Fixes#52921
Fields can declare a longer 'info' text rendered as an (i) tooltip next to
the label in both the panel and the modal; honcho uses it to spell out the
session strategy, write frequency, and recall mode semantics. The modal
description claimed 'the active profile' without saying which — show the
active gateway profile name.
Providers with behavior beyond fields (validate a server, start a local
instance, link a CLI profile) previously had no mount besides forking the
panel. Let the schema declare actions and one generic endpoint run them:
ProviderAction on ProviderConfigSchema, POST
/api/memory/providers/{name}/actions/{action} dispatching to
ACTION_HANDLERS in the plugin's config_actions.py (path-loaded and
import-light, like the schema), profile-scoped and off the event loop.
Handlers get the submitted values dict, return a JSON-able result, and
raise ValueError for user-facing 400s. The panel renders declared actions
as generic buttons beside Save. No bundled provider declares actions yet.
The settings page follows the desktop's active-profile switcher, but the
provider config calls didn't: no profileScoped() on the client and no
profile param on the backend, so a multi-profile desktop edited the serving
process's config while every surrounding card showed the selected profile's.
Accept ?profile= on both endpoints and resolve inside _profile_scope (the
skills/toolsets contract), spread profileScoped() into the two client calls,
and key the schema cache on the resolved config_schema.py path instead of
the provider name — user-installed plugins are per-profile, so one profile's
lookup must never answer for another's.
Replace the hand-rolled ensurepip bootstrap (and five other one-off
pip-install code paths) with hermes_cli.tools_config._pip_install, which
prefers the bundled uv (fast, needs no pip in the venv), falls back to
python -m pip, and bootstraps pip via ensurepip only when missing.
Sites unified:
- hermes_cli/setup.py: _install_neutts_deps, _install_kittentts_deps,
modal SDK install, daytona SDK install
- hermes_cli/memory_setup.py: memory-plugin pip deps (previously dead-ended
when uv AND pip binaries were both absent)
- hermes_cli/dingtalk_auth.py: qrcode auto-install (previously invoked
'python -m uv' which is not how uv ships)
- agent/lsp/install.py: --target LSP server installs
- plugins/google_meet/cli.py, plugins/platforms/matrix/adapter.py,
plugins/platforms/google_chat/oauth.py, plugins/memory/honcho/cli.py
Tests updated to assert the ladder behavior (uv-first, pip fallback,
ensurepip bootstrap) instead of the removed bespoke branches.
ruff check --fix --select F541 . on current main. Pure prefix removals;
adjacent-string concatenations keep the f only on interpolating fragments.
No string content or live placeholder altered.
Multi-line comment blocks and JSDoc-style headers across the provider
config surface compress to one line each; the why lives in commit messages
and docstrings, not comment essays.
A syntax error in a provider's config_schema.py was cached as 'no schema'
until process restart, rendering a silent empty panel even after the file
was fixed. Cache only successful loads and genuine file-absence.
Each provider now declares its config surface in config_schema.py inside
its own plugin dir (plugins/memory/<name>/), loaded by file path like the
plugins themselves so plugin __init__ imports never reach the web server.
hermes_cli/memory_providers.py is gone; the shared field primitives and
loader live in plugins/memory/config_schema.py, and the schema tests move
to tests/plugins/memory/ alongside the other per-plugin suites.
