fix(honcho): plugin drift overhaul -- observation config, chunking, setup wizard, docs, dead code cleanup from migration by erosika · Pull Request #5045 · NousResearch/hermes-agent

erosika · 2026-04-04T15:00:42Z

Summary

Honcho memory plugin overhaul addressing integration drift and community feedback since the v0.7 plugin migration. Fixes longstanding config mismatches, adds missing API limit handling, replaces opaque presets with granular per-peer observation, fills documentation gaps, and now reads profile data directly from peer APIs when session-scoped context comes back empty.

Fixes

base_url/baseUrl mismatch: client reads both conventions, schema normalized to camelCase, CLI cleans up legacy keys on setup. Fixes self-hosted setups that silently failed.
Setup command conflation: hermes honcho setup and hermes memory setup now converge to the same full wizard via post_setup() hook. No more redirect to a 2-field generic form.
SOUL.md aiPeer override: removed -- persona content should not silently override identity config.
Local auth leak: cloud API keys no longer sent to local Honcho instances. Only uses key if the host block explicitly sets apiKey.
Dead code: removed memoryMode / peer_memory_modes (never enforced by MemoryManager), linkedHosts (never called), suppress flags (set but never read). cmd_mode repurposed from dead memoryMode to recallMode.
Picker hint: shows "API key / local" instead of "requires API key" for providers supporting both.
Peer-card/profile reads: honcho_profile, search, and prefetch paths now fall back to direct peer APIs (get_card() / context() / representation()) instead of relying only on session-scoped context, which could return empty profile data in messaging flows.

New features

Message chunking: messages exceeding the Honcho API limit (default 25k) are split at paragraph/sentence/word boundaries with [continued] markers. Configurable via messageMaxChars.
Dialectic input guard: queries exceeding 10k truncated at word boundary before peer.chat(). Configurable via dialecticMaxInputChars.
Granular per-peer observation: replaces observationMode string presets with per-peer observeMe/observeOthers booleans mapping 1:1 to Honcho's SessionPeerConfig. Legacy presets still work as shorthands. Default: all observations on.
Server-side observation sync: get_peer_configuration() called after add_peers on session init -- dashboard changes picked up automatically.
dialecticDynamic toggle: configurable auto-bump of reasoning level by query length (default true). Set false for fixed level.
Setup wizard rewrite: cloud/local deployment picker, explicit AI peer prompt, observation mode, auto-enables memory.provider: honcho in config.yaml.
Profile clone improvements: clone_honcho_for_profile and cmd_enable now inherit observation, dialecticDynamic, messageMaxChars, dialecticMaxInputChars.

Docs

plugins/memory/honcho/README.md: full config reference (30+ keys, defaults, scope, precedence, granular observation docs, per-profile examples)
memory-providers.md: Honcho section rewritten with self-hosted support, collapsible config table, minimal JSON examples (cloud + local), multi-profile section with --clone and hermes honcho sync explanation
profiles.md: tip linking to Honcho multi-profile docs
New optional skill autonomous-ai-agents/honcho: setup, architecture, multi-profile, observation, all 16 CLI commands, troubleshooting

Tests

Fixed 2 pre-existing TestResolveConfigPath failures (filesystem isolation)
9 new tests: message chunking (paragraph/sentence/word splits, continuation prefix, large messages) + dialectic input guard
Added regression coverage for direct peer-card / peer-context lookup behavior
Removed 14 dead tests for removed memoryMode / linkedHosts code

Close on merge

PRs:

fix(honcho): sync aiPeer with SOUL.md identity at runtime #1969 fix(honcho): sync aiPeer with SOUL.md identity at runtime
fix(honcho): support local baseUrl-only configuration #2645 fix(honcho): support local baseUrl-only configuration
fix(honcho): enforce contextTokens budget on prefetched Honcho context #3265 fix(honcho): enforce contextTokens budget on prefetched Honcho context
feat(honcho): configurable prefetch cadence, injection toggles, and reasoning cap #3425 feat(honcho): configurable prefetch cadence, injection toggles, and reasoning cap
fix(honcho): skip Honcho writes for cron sessions #4053 fix(honcho): skip Honcho writes for cron sessions
fix(honcho): profile isolation and peer naming #4632 fix(honcho): profile isolation and peer naming
fix(honcho): port profile isolation fixes to plugin architecture #4642 fix(honcho): port profile isolation fixes to plugin architecture
fix(honcho): respect nested host aiPeer config in SOUL.md sync #5061 fix(honcho): respect nested host aiPeer config in SOUL.md sync
fix(honcho): read peer cards from direct peer APIs #5137 fix(honcho): read peer cards from direct peer APIs
fix(honcho): schema key mismatch breaks self-hosted Honcho config #5184 fix(honcho): schema key mismatch breaks self-hosted Honcho config
fix(honcho): use self-observation for peer_card queries #5231 fix(honcho): use self-observation for peer_card queries

Issues:

[Bug]: Local Honcho "base_url" instead of api key does not work #2613 [Bug]: Local Honcho "base_url" instead of api key does not work
feat(honcho): configurable prefetch cadence, injection granularity, and cost-efficient defaults #3422 feat(honcho): configurable prefetch cadence, injection granularity, and cost-efficient defaults
[Bug]: Invalid Honcho API key from local host??? #3555 [Bug]: Invalid Honcho API key from local host???
Self-hosted Honcho setup/docs drift: wizard requires API key and current code path differs from docs #4296 Self-hosted Honcho setup/docs drift: wizard requires API key and current code path differs from docs

erosika · 2026-04-04T15:55:54Z

CI failures are pre-existing and unrelated to this PR:

Supply chain audit: scan passed but the bot can't post comments on fork PRs (Resource not accessible by integration). Permissions issue, not a finding.
Tests (11 failures): all in test_codex_execution_paths (1) and test_skill_manager_tool (10). None touch honcho, memory, or any file in this diff. 7884 tests passed including all honcho plugin tests.

…ervation, dialectic guards Incorporates Honcho memory UX improvements from upstream PR NousResearch#5045: Observation system: - Default observation flipped: unified -> directional (all 4 bools true) - Granular per-peer observation booleans replace string-only mode - Server-side observation config sync after add_peers() - Observation presets (directional, unified) kept as shorthands - Per-peer override via 'observation' JSON object in honcho.json Message chunking: - _chunk_message() splits at paragraph/sentence/word boundaries - Replaces hard [:4000] truncation in sync_turn() - Configurable via messageMaxChars (default 25000) - Continuation chunks prefixed with '[continued]' Dialectic controls: - dialecticDynamic toggle (false = no auto-bump, cost control) - dialecticMaxInputChars guard (default 10000, truncates long queries) - Applied in both session.py dialectic_query() and __init__.py tool handler Config schema: - New fields exposed via get_config_schema() for hermes memory setup - observation, messageMaxChars, dialecticDynamic, dialecticMaxInputChars Removals (dead code from PR NousResearch#5045): - memory_mode, peer_memory_modes, linked_hosts already removed in prior commits - String-based observation mode logic replaced with boolean-based

…up wizard, tests, docs, skill Completes adoption of erosika's Honcho plugin overhaul (PR NousResearch#5045). Previous commit adopted core plugin files (__init__, client, session). This commit adds: - plugins/memory/honcho/cli.py: full CLI with setup wizard rewrite (cloud/local picker, observation mode, memoryMode->recallMode migration) - hermes_cli/memory_setup.py: dependency install improvements - tests/honcho_plugin/test_session.py: 9 new tests (chunking, dialectic guard) - tests/honcho_plugin/test_client.py: config resolution fixes - tests/honcho_plugin/test_async_memory.py: dead test cleanup - plugins/memory/honcho/README.md: full config reference (30+ keys) - optional-skills/autonomous-ai-agents/honcho/SKILL.md: new honcho skill

- client.py reads both baseUrl and base_url from config (backward compat) - schema declaration changed from base_url to baseUrl (correct JSON convention) - fix 2 TestResolveConfigPath failures: mock Path.home() to isolate from real ~/.hermes/honcho.json during test - add messageMaxChars (default 25k) and dialecticMaxInputChars (default 10k) config fields for Honcho API limits - granular per-peer observation config: observation object with user/ai observeMe/observeOthers booleans, replacing string presets - server-side observation sync via get_peer_configuration() on session init Resolves: spec H1 (base_url/baseUrl mismatch)

Setup wizard: - cloud/local deployment picker with smart API key handling - explicit AI peer name prompt (never derived from SOUL.md) - observation mode config, auto-enables memory.provider in config.yaml - post_setup() hook: both hermes honcho setup and hermes memory setup converge to the same full wizard via cmd_setup_provider() - hardcode memoryMode: hybrid (honcho-only was dead code) - strip emojis from CLI output - clean up legacy base_url snake_case key on setup Memory plugin: - message chunking: split at paragraph/sentence/word boundaries with [continued] markers when exceeding messageMaxChars (default 25k) - removed SOUL.md aiPeer sync (persona content != identity config) - removed dead _sanitize_peer_name + import re - schema key fixed: base_url -> baseUrl Picker: - "API key / local" hint for providers with both secret and non-secret fields - removed active indicator per feedback Resolves: spec C1, C2, H2, M2

session.py: - per-peer observation booleans replace mode string branching - SessionPeerConfig set from granular user/ai observeMe/observeOthers - server-side config sync: get_peer_configuration() after add_peers so Honcho dashboard changes are picked up on session init - dialectic input guard: truncate queries exceeding dialecticMaxInputChars (default 10k) at word boundary - routing logic keys off ai_observe_others instead of mode string tests: - 8 chunking tests (paragraph/sentence/word split, continuation prefix, empty, large message, exact limit) - 1 dialectic input guard test (long query truncated) Default observation mode changed: directional (all on) replaces unified

README.md: full config reference with 30+ keys, all defaults, scope, precedence rules, granular observation docs, example config, CLI commands memory-providers.md: - honcho section: self-hosted support, API key link to app.honcho.dev - setup wizard command, config reference link to GitHub README - multi-profile section with --clone and hermes honcho sync optional skill (autonomous-ai-agents/honcho): - setup (cloud/local), architecture (peers, observation, sessions) - multi-profile with --clone, per-profile observation override - tool reference with cost/latency guidance - config reference, cost-awareness advanced settings - troubleshooting for 5 common issues

Add dialecticDynamic (bool, default true) to control whether reasoning level auto-bumps based on query length. When false, always uses dialecticReasoningLevel as-is. Documented in README and skill.

Dead code removed: - memoryMode / memory_mode / peer_memory_modes / peer_memory_mode() / _resolve_memory_mode() -- C1: honcho-only mode was never enforced, MemoryManager always runs both providers. Setup hardcodes hybrid. - linkedHosts / linked_hosts / get_linked_workspaces() -- never called from session, cli, or init. Pure dead weight. - B2 suppress flags (_suppress_memory, _suppress_user_profile) -- set but never read by anything. - 14 dead tests for the above. Profile clone key lists updated: - clone_honcho_for_profile + cmd_enable now inherit observation, dialecticDynamic, messageMaxChars, dialecticMaxInputChars. - Removed memoryMode from clone lists. Docs: - memory-providers.md: collapsible config reference + full multi-profile example with <details>, expanded hermes honcho sync description - profiles.md: tip linking to Honcho multi-profile section - README: per-profile observation example, removed memoryMode + linkedHosts

…e, scope comment - _chunk_message: consolidate budget/effective_limit into single var - cli.py: remove all memoryMode writes and display references - cmd_mode: repurposed from dead memoryMode to recallMode (hybrid/context/tools) - cmd_status: replace memory mode line with observation config display - session.py: add comment noting observation sync is manager-scoped not per-session - session.py: fix stale docstring reference to memory_mode

- fetch honcho_profile card data from peer.get_card()/peer.card() - use peer-level context/representation lookups for search and prefetch - add regression tests covering direct peer lookup behavior

…etup wizard, docs, dead code cleanup Salvaged from PR #5045 by erosika. - Replace memoryMode/peer_memory_modes with granular per-peer observation config - Add message chunking for Honcho API limits (25k chars default) - Add dialectic input guard (10k chars default) - Add dialecticDynamic toggle for reasoning level auto-bump - Rewrite setup wizard with cloud/local deployment picker - Switch peer card/profile/search from session.context() to direct peer APIs - Add server-side observation sync via get_peer_configuration() - Fix base_url/baseUrl config mismatch for self-hosted setups - Fix local auth leak (cloud API keys no longer sent to local instances) - Remove dead code: memoryMode, peer_memory_modes, linkedHosts, suppress flags, SOUL.md aiPeer sync - Add post_setup hook to memory_setup.py for provider-specific setup wizards - Comprehensive README rewrite with full config reference - New optional skill: autonomous-ai-agents/honcho - Expanded memory-providers.md with multi-profile docs - 9 new tests (chunking, dialectic guard, peer lookups), 14 dead tests removed - Fix 2 pre-existing TestResolveConfigPath filesystem isolation failures

teknium1 · 2026-04-05T19:33:37Z

Merged via PR #5295. Your plugin-internal changes (observation config, message chunking, dialectic guard, direct peer APIs, setup wizard rewrite, dead code cleanup, docs) were cherry-picked with your authorship preserved.

On top of your work, we made a few architectural changes:

Plugin CLI registration system: Instead of keeping the hardcoded honcho argparse block in main.py, we added a register_cli(subparser) convention so plugins register their own CLI commands. The 95-line honcho block was moved entirely into plugins/memory/honcho/cli.py. Future plugins can add CLI commands without touching core files.
Active-provider gating: Plugin CLI commands only appear when the corresponding memory provider is active (memory.provider in config.yaml). If honcho isn't the active provider, hermes honcho doesn't show up.
Migration guard for observation default: Your PR changed the default from "unified" to "directional". We added a guard so existing configs without explicit observationMode keep "unified" (old default), while new installs get "directional".
Stripped unrelated changes: matrix.py, skill_manager_tool.py, and browser test changes were removed from the salvage since they weren't related to the honcho overhaul.
hermes honcho setup redirects to hermes memory setup: Setup goes through the unified memory provider path. The post_setup hook you added lets the honcho wizard run after provider selection.

Thanks for the thorough work — this closes 11 PRs and 4 issues in one shot.

…etup wizard, docs, dead code cleanup Salvaged from PR #5045 by erosika. - Replace memoryMode/peer_memory_modes with granular per-peer observation config - Add message chunking for Honcho API limits (25k chars default) - Add dialectic input guard (10k chars default) - Add dialecticDynamic toggle for reasoning level auto-bump - Rewrite setup wizard with cloud/local deployment picker - Switch peer card/profile/search from session.context() to direct peer APIs - Add server-side observation sync via get_peer_configuration() - Fix base_url/baseUrl config mismatch for self-hosted setups - Fix local auth leak (cloud API keys no longer sent to local instances) - Remove dead code: memoryMode, peer_memory_modes, linkedHosts, suppress flags, SOUL.md aiPeer sync - Add post_setup hook to memory_setup.py for provider-specific setup wizards - Comprehensive README rewrite with full config reference - New optional skill: autonomous-ai-agents/honcho - Expanded memory-providers.md with multi-profile docs - 9 new tests (chunking, dialectic guard, peer lookups), 14 dead tests removed - Fix 2 pre-existing TestResolveConfigPath filesystem isolation failures

teknium1 · 2026-04-05T19:34:26Z

Closed — merged via PR #5295 with your authorship preserved on the honcho overhaul commit. See the comment above for details on the architectural additions we made on top. Thanks for the thorough work!

…etup wizard, docs, dead code cleanup Salvaged from PR NousResearch#5045 by erosika. - Replace memoryMode/peer_memory_modes with granular per-peer observation config - Add message chunking for Honcho API limits (25k chars default) - Add dialectic input guard (10k chars default) - Add dialecticDynamic toggle for reasoning level auto-bump - Rewrite setup wizard with cloud/local deployment picker - Switch peer card/profile/search from session.context() to direct peer APIs - Add server-side observation sync via get_peer_configuration() - Fix base_url/baseUrl config mismatch for self-hosted setups - Fix local auth leak (cloud API keys no longer sent to local instances) - Remove dead code: memoryMode, peer_memory_modes, linkedHosts, suppress flags, SOUL.md aiPeer sync - Add post_setup hook to memory_setup.py for provider-specific setup wizards - Comprehensive README rewrite with full config reference - New optional skill: autonomous-ai-agents/honcho - Expanded memory-providers.md with multi-profile docs - 9 new tests (chunking, dialectic guard, peer lookups), 14 dead tests removed - Fix 2 pre-existing TestResolveConfigPath filesystem isolation failures

…ervation, dialectic guards

…etup wizard, docs, dead code cleanup Salvaged from PR NousResearch#5045 by erosika. - Replace memoryMode/peer_memory_modes with granular per-peer observation config - Add message chunking for Honcho API limits (25k chars default) - Add dialectic input guard (10k chars default) - Add dialecticDynamic toggle for reasoning level auto-bump - Rewrite setup wizard with cloud/local deployment picker - Switch peer card/profile/search from session.context() to direct peer APIs - Add server-side observation sync via get_peer_configuration() - Fix base_url/baseUrl config mismatch for self-hosted setups - Fix local auth leak (cloud API keys no longer sent to local instances) - Remove dead code: memoryMode, peer_memory_modes, linkedHosts, suppress flags, SOUL.md aiPeer sync - Add post_setup hook to memory_setup.py for provider-specific setup wizards - Comprehensive README rewrite with full config reference - New optional skill: autonomous-ai-agents/honcho - Expanded memory-providers.md with multi-profile docs - 9 new tests (chunking, dialectic guard, peer lookups), 14 dead tests removed - Fix 2 pre-existing TestResolveConfigPath filesystem isolation failures

erosika changed the title ~~feat(honcho): plugin overhaul -- observation config, chunking, setup wizard, docs~~ fix(honcho): plugin drift overhaul -- observation config, chunking, setup wizard, docs Apr 4, 2026

erosika changed the title ~~fix(honcho): plugin drift overhaul -- observation config, chunking, setup wizard, docs~~ fix(honcho): plugin drift overhaul -- observation config, chunking, setup wizard, docs, dead code cleanup Apr 4, 2026

erosika changed the title ~~fix(honcho): plugin drift overhaul -- observation config, chunking, setup wizard, docs, dead code cleanup~~ fix(honcho): plugin drift overhaul -- observation config, chunking, setup wizard, docs, dead code cleanup from migration Apr 4, 2026

erosika changed the title ~~fix(honcho): plugin drift overhaul -- observation config, chunking, setup wizard, docs, dead code cleanup from migration~~ feat(honcho): plugin overhaul -- observation config, chunking, setup wizard, docs Apr 4, 2026

erosika changed the title ~~feat(honcho): plugin overhaul -- observation config, chunking, setup wizard, docs~~ fix(honcho): plugin drift overhaul -- observation config, chunking, setup wizard, docs, dead code cleanup from migration Apr 4, 2026

erosika force-pushed the eri/honcho-plugin-overhaul branch from cb85697 to 7c982ba Compare April 5, 2026 14:14

erosika and others added 11 commits April 5, 2026 10:23

feat(honcho): configurable dialecticDynamic toggle for reasoning level

8820d4e

Add dialecticDynamic (bool, default true) to control whether reasoning level auto-bumps based on query length. When false, always uses dialecticReasoningLevel as-is. Documented in README and skill.

docs: move honcho tip to end of clone section in profiles

38625b3

docs(honcho): add config path resolution, minimal JSON examples

dc4de3b

docs(honcho): full CLI command reference in skill

b27b408

fix(honcho): read peer cards from direct peer APIs

ea95cc0

- fetch honcho_profile card data from peer.get_card()/peer.card() - use peer-level context/representation lookups for search and prefetch - add regression tests covering direct peer lookup behavior

erosika force-pushed the eri/honcho-plugin-overhaul branch from 7c982ba to ea95cc0 Compare April 5, 2026 14:23

fix: address CI regressions on 5045 branch

79a2cbf

This was referenced Apr 5, 2026

fix(hindsight): fix hindsight memory plugin and memory provider setup #5044

Closed

bug: hidden MemoryProvider prefetch is framed as user-visible conversation #5289

Closed

teknium1 mentioned this pull request Apr 5, 2026

fix(honcho): plugin drift overhaul + feat(plugins): CLI registration system #5295

Merged

teknium1 closed this Apr 5, 2026

zebster-cmd added a commit to zebster-cmd/hermes-agent that referenced this pull request Apr 18, 2026

feat(honcho): adopt PR NousResearch#5045 improvements — chunking, obs…

7ede621

…ervation, dialectic guards

teknium1 mentioned this pull request Apr 27, 2026

fix(honcho): use self-observation for peer_card queries #5231

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix(honcho): plugin drift overhaul -- observation config, chunking, setup wizard, docs, dead code cleanup from migration#5045

fix(honcho): plugin drift overhaul -- observation config, chunking, setup wizard, docs, dead code cleanup from migration#5045
erosika wants to merge 12 commits into
NousResearch:mainfrom
erosika:eri/honcho-plugin-overhaul

erosika commented Apr 4, 2026 •

edited

Loading

Uh oh!

erosika commented Apr 4, 2026

Uh oh!

teknium1 commented Apr 5, 2026

Uh oh!

teknium1 commented Apr 5, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

erosika commented Apr 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Fixes

New features

Docs

Tests

Close on merge

Uh oh!

erosika commented Apr 4, 2026

Uh oh!

teknium1 commented Apr 5, 2026

Uh oh!

teknium1 commented Apr 5, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

erosika commented Apr 4, 2026 •

edited

Loading