{ "schemaVersion": "1.0", "item": { "slug": "honcho-memory", "name": "Honcho Memory", "source": "tencent", "type": "skill", "category": "通讯协作", "sourceUrl": "https://clawhub.ai/JKILLR/honcho-memory", "canonicalUrl": "https://clawhub.ai/JKILLR/honcho-memory", "targetPlatform": "OpenClaw" }, "install": { "downloadMode": "manual_only", "downloadUrl": "/downloads/honcho-memory", "sourceDownloadUrl": "https://wry-manatee-359.convex.site/api/v1/download?slug=honcho-memory", "sourcePlatform": "tencent", "targetPlatform": "OpenClaw", "installMethod": "Manual import", "extraction": "Extract archive", "prerequisites": [ "OpenClaw" ], "packageFormat": "ZIP package", "includedAssets": [ "SKILL.md", "references/api-reference.md", "references/architecture.md", "scripts/feed.py", "scripts/generate-context.py", "scripts/query.py" ], "primaryDoc": "SKILL.md", "quickSetup": [ "Open the source page and confirm the package flow manually.", "Review SKILL.md if you can obtain the files.", "Treat this source as manual setup until the download is verified." ], "agentAssist": { "summary": "Use the source page and any available docs to guide the install because the item currently does not return a direct package file.", "steps": [ "Open the source page via Open source listing.", "If you can obtain the package, extract it into a folder your agent can access.", "Paste one of the prompts below and point your agent at the source page and extracted files." ], "prompts": [ { "label": "New install", "body": "I tried to install a skill package from Yavira, but the item currently does not return a direct package file. Inspect the source page and any extracted docs, then tell me what you can confirm and any manual steps still required." }, { "label": "Upgrade existing", "body": "I tried to upgrade a skill package from Yavira, but the item currently does not return a direct package file. Compare the source page and any extracted docs with my current installation, then summarize what changed and what manual follow-up I still need." } ] }, "sourceHealth": { "source": "tencent", "slug": "honcho-memory", "status": "source_issue", "reason": "not_found", "recommendedAction": "review_source", "checkedAt": "2026-04-30T06:02:24.133Z", "expiresAt": "2026-05-01T06:02:24.133Z", "httpStatus": 404, "finalUrl": "https://wry-manatee-359.convex.site/api/v1/download?slug=honcho-memory", "contentType": "text/plain", "probeMethod": "head", "details": { "probeUrl": "https://wry-manatee-359.convex.site/api/v1/download?slug=honcho-memory", "contentDisposition": null, "redirectLocation": null, "bodySnippet": null, "slug": "honcho-memory" }, "scope": "item", "summary": "Known item issue.", "detail": "This item's current download entry is known to bounce back to a listing or homepage instead of returning a package file.", "primaryActionLabel": "Open source listing", "primaryActionHref": "https://clawhub.ai/JKILLR/honcho-memory" }, "validation": { "installChecklist": [ "Open the source listing and confirm there is a real package or setup artifact available.", "Review SKILL.md before asking your agent to continue.", "Treat this source as manual setup until the upstream download flow is fixed." ], "postInstallChecks": [ "Confirm the extracted package includes the expected docs or setup files.", "Validate the skill or prompts are available in your target agent workspace.", "Capture any manual follow-up steps the agent could not complete." ] }, "downloadPageUrl": "https://openagent3.xyz/downloads/honcho-memory", "agentPageUrl": "https://openagent3.xyz/skills/honcho-memory/agent", "manifestUrl": "https://openagent3.xyz/skills/honcho-memory/agent.json", "briefUrl": "https://openagent3.xyz/skills/honcho-memory/agent.md" }, "agentAssist": { "summary": "Use the source page and any available docs to guide the install because the item currently does not return a direct package file.", "steps": [ "Open the source page via Open source listing.", "If you can obtain the package, extract it into a folder your agent can access.", "Paste one of the prompts below and point your agent at the source page and extracted files." ], "prompts": [ { "label": "New install", "body": "I tried to install a skill package from Yavira, but the item currently does not return a direct package file. Inspect the source page and any extracted docs, then tell me what you can confirm and any manual steps still required." }, { "label": "Upgrade existing", "body": "I tried to upgrade a skill package from Yavira, but the item currently does not return a direct package file. Compare the source page and any extracted docs with my current installation, then summarize what changed and what manual follow-up I still need." } ] }, "documentation": { "source": "clawhub", "primaryDoc": "SKILL.md", "sections": [ { "title": "Honcho Memory", "body": "Production memory system for AI agents, built on Honcho by Plastic Labs. Developed and refined running a fleet of 6+ agents in production with 1000+ messages fed through the reasoning engine.\n\nThis isn't a wrapper or setup guide. It's a complete memory pipeline: feed conversations in, reason over them automatically, generate token-budgeted context files, and query on-demand when agents need to recall something they've never seen in their current session." }, { "title": "The Problem This Solves", "body": "Agent memory is broken in three ways:\n\nCompaction amnesia. Long sessions get compacted. Context vanishes. Your agent forgets decisions made 2 hours ago.\nIsolated session blindness. Cron jobs and background tasks spin up fresh sessions with zero conversation history. They operate without context.\nSingle-agent silos. In multi-agent setups, agents can't access what other agents learned. Knowledge stays trapped in individual sessions.\n\nHoncho Memory solves all three. Every agent feeds into a shared reasoning engine. Every session (main, cron, isolated) loads reasoned context at startup. Nothing gets lost." }, { "title": "How It Works", "body": "Most agent memory is embedding search: store text chunks, retrieve similar chunks later. That's a library with a search bar.\n\nHoncho is a reasoning engine. It processes conversations through Neuromancer and builds deductive and inductive observations about every person and agent it tracks. It doesn't retrieve what was said. It reasons about what it means.\n\nFeed it 3 weeks of multi-agent conversations and ask \"What should the social media agent focus on this week?\" It won't dump old messages. It will synthesize: \"Based on recent engagement patterns, sports/entertainment replies are outperforming crypto content 3:1. The user shifted priority toward brand-building over direct promotion last Tuesday.\"" }, { "title": "Architecture", "body": "WRITE PATH\nConversations ──→ feed.py ──→ Honcho API ──→ Neuromancer ──→ Observations\n (all agents) │ (async reasoning) │\n │ ↓\n │ Representations\n │ (per peer)\n │ │\n │ READ PATH │\n │ ┌─────────────────────────────┐ │\n │ │ generate-context.py (cron) │←───────────┘\n │ │ Token-budgeted files that │\n │ │ survive compaction │\n │ └──────────┬──────────────────┘\n │ │\n │ ↓\n │ HONCHO-CONTEXT.md (shared)\n │ agents/*/HONCHO-CONTEXT.md (per-agent)\n │ │\n │ ↓\n │ Every session loads reasoned context\n │ (main, cron, isolated, sub-agent)\n │\n │ ┌─────────────────────────────┐\n └───→│ query.py (on-demand) │←── Agent mid-conversation\n │ \"What did we decide about │ needs to recall something\n │ pricing last week?\" │\n └─────────────────────────────┘" }, { "title": "Compaction-Proof Memory", "body": "Context files are workspace files. They survive compaction, session resets, and gateway restarts. Every session starts with reasoned context automatically, not just the last few messages." }, { "title": "Cross-Agent Memory", "body": "All agents feed into the same Honcho workspace. Agent A learns something on Monday. Agent B can access that understanding on Tuesday through shared context generation. No manual syncing, no copy-pasting between workspaces." }, { "title": "Token-Budgeted Context", "body": "Generated context files are budget-controlled (~1000-1500 tokens shared, ~500-800 per agent). No prompt bloat. Honcho distills weeks of conversations into dense, relevant context that fits cleanly in the system prompt." }, { "title": "Reasoning Levels", "body": "On-demand queries support five reasoning levels: minimal, low, medium, high, max. Quick lookups stay cheap. Deep analysis gets thorough treatment." }, { "title": "Honcho-First, Files as Fallback", "body": "Designed to complement existing file-based memory (MEMORY.md, memory/*.md), not replace it. Honcho handles reasoned recall. Local files handle raw logs and structured data. If Honcho is unavailable, file-based memory still works." }, { "title": "Prerequisites", "body": "Python 3.10+\nHoncho account + API key (app.honcho.dev)\npip install honcho-ai" }, { "title": "1. Install SDK", "body": "pip install honcho-ai" }, { "title": "2. Configure credentials", "body": "Create ~/.config/honcho/credentials.json:\n\n{\n \"api_key\": \"your-honcho-api-key\",\n \"workspace_id\": \"your-workspace-name\"\n}" }, { "title": "3. Initialize workspace and peers", "body": "python3 scripts/setup.py --workspace \"my-workspace\" --peers \"user,agent1,agent2\"\n\nCreates the workspace and registers every entity (users and agents) that Honcho will track and build representations for." }, { "title": "Feed conversations", "body": "python3 scripts/feed.py\n\nAuto-discovers active agent session transcripts, attributes messages to correct peers, tracks sync position to avoid duplicates, feeds in batches. Run after sessions or on a cron schedule." }, { "title": "Query on-demand", "body": "When an agent needs context it doesn't have:\n\npython3 scripts/query.py \"What was the decision on pricing?\"\npython3 scripts/query.py \"What are the user's priorities right now?\" --peer user-name\npython3 scripts/query.py \"Full analysis of this week's performance\" --level high" }, { "title": "Generate context files", "body": "python3 scripts/generate-context.py # all agents\npython3 scripts/generate-context.py --agent agent-name # specific agent\npython3 scripts/generate-context.py --query \"topic\" # on-demand deep query\n\nProduces:\n\nHONCHO-CONTEXT.md in workspace root (shared across all agents)\nHONCHO-CONTEXT.md in each agent's workspace (agent-specific)" }, { "title": "Check sync status", "body": "python3 scripts/feed.py --status" }, { "title": "Cron Automation", "body": "Run hourly during active hours to keep memory fresh:\n\n0 9-23 * * * cd /workspace && python3 scripts/feed.py && python3 scripts/generate-context.py\n\nOpenClaw cron tool config:\n\n{\n \"name\": \"honcho-sync\",\n \"schedule\": {\"kind\": \"cron\", \"expr\": \"0 9-23 * * *\", \"tz\": \"America/Los_Angeles\"},\n \"sessionTarget\": \"isolated\",\n \"payload\": {\n \"kind\": \"agentTurn\",\n \"message\": \"Run: cd /workspace && source .venv/bin/activate && python3 scripts/feed.py && python3 scripts/generate-context.py. Report summary.\",\n \"timeoutSeconds\": 180\n },\n \"delivery\": {\"mode\": \"none\"}\n}" }, { "title": "Multi-Agent Setup", "body": "For fleets with multiple agents, each agent gets:\n\nIts own peer in Honcho (tracks what that agent said and learned)\nShared context from the root HONCHO-CONTEXT.md (cross-agent awareness)\nAgent-specific context from its workspace HONCHO-CONTEXT.md (role-specific memory)\n\nThe feed script handles attribution automatically. User messages go to the human peer. Each agent's responses go to that agent's peer. Honcho reasons across all of them.\n\nExample: 6-agent fleet\n\nMain agent, social media agent, content agent, community agent, news scanner, expansion agent\nAll feed into one Honcho workspace\nEach gets tailored context files reflecting their role and recent activity\nThe main agent's context includes a cross-agent brief summarizing what every other agent did" }, { "title": "Integration with Agent Prompts", "body": "Add to each agent's system prompt or operating instructions:\n\n## Memory\n- At session start: read HONCHO-CONTEXT.md for reasoned context from all conversations\n- If you don't remember something or need context you don't have:\n python3 scripts/query.py \"your question\" --peer user-name\n- Honcho reasons over ALL agent conversations. It likely has the answer.\n- Before saying \"I don't know\" about past decisions, projects, or conversations: query Honcho first." }, { "title": "Use Cases", "body": "Cron job needs context: Generated context files give every isolated session full awareness of recent activity, decisions, and priorities without carrying conversation history.\nAgent forgot a decision: On-demand query pulls reasoned context about any topic from weeks of conversations.\nNew agent joins the fleet: Reads shared HONCHO-CONTEXT.md and immediately understands the user, active projects, and current priorities.\nPost-compaction recovery: Context files survive compaction. Agent reloads and continues without losing the thread.\nCross-agent coordination: Social media agent needs to know what the content agent published. Shared context provides this automatically." }, { "title": "How Honcho's Reasoning Works", "body": "Honcho's data model: Workspaces → Peers → Sessions → Messages. Workspaces isolate environments. Peers are any entity that persists and changes over time (users, agents). Sessions are conversation threads between peers. Messages are units of data that trigger reasoning.\n\nWhen messages are written, Honcho doesn't just store them. Custom reasoning models (Neuromancer) perform formal logical reasoning in the background to generate conclusions about each peer. These conclusions compound into representations that evolve with every new interaction. When you query, you get reasoned context derived from representations, not raw message retrieval.\n\nObservations: Deductive and inductive conclusions drawn from conversations\nRepresentations: Synthesized, evolving understanding of each peer\nContext retrieval: Token-budgeted, combining representations, recent activity, and reasoned observations\n\nFor full details on Honcho's architecture: docs.honcho.dev\n\nCost: ~$2/million tokens ingested. Storage and retrieval free. Most multi-agent setups run $4-6/month." }, { "title": "References", "body": "references/api-reference.md — Full Honcho SDK API documentation\nreferences/architecture.md — Integration patterns and data flow diagrams\nHoncho docs\nPlastic Labs" } ], "body": "Honcho Memory\n\nProduction memory system for AI agents, built on Honcho by Plastic Labs. Developed and refined running a fleet of 6+ agents in production with 1000+ messages fed through the reasoning engine.\n\nThis isn't a wrapper or setup guide. It's a complete memory pipeline: feed conversations in, reason over them automatically, generate token-budgeted context files, and query on-demand when agents need to recall something they've never seen in their current session.\n\nThe Problem This Solves\n\nAgent memory is broken in three ways:\n\nCompaction amnesia. Long sessions get compacted. Context vanishes. Your agent forgets decisions made 2 hours ago.\nIsolated session blindness. Cron jobs and background tasks spin up fresh sessions with zero conversation history. They operate without context.\nSingle-agent silos. In multi-agent setups, agents can't access what other agents learned. Knowledge stays trapped in individual sessions.\n\nHoncho Memory solves all three. Every agent feeds into a shared reasoning engine. Every session (main, cron, isolated) loads reasoned context at startup. Nothing gets lost.\n\nHow It Works\n\nMost agent memory is embedding search: store text chunks, retrieve similar chunks later. That's a library with a search bar.\n\nHoncho is a reasoning engine. It processes conversations through Neuromancer and builds deductive and inductive observations about every person and agent it tracks. It doesn't retrieve what was said. It reasons about what it means.\n\nFeed it 3 weeks of multi-agent conversations and ask \"What should the social media agent focus on this week?\" It won't dump old messages. It will synthesize: \"Based on recent engagement patterns, sports/entertainment replies are outperforming crypto content 3:1. The user shifted priority toward brand-building over direct promotion last Tuesday.\"\n\nArchitecture\n WRITE PATH\nConversations ──→ feed.py ──→ Honcho API ──→ Neuromancer ──→ Observations\n (all agents) │ (async reasoning) │\n │ ↓\n │ Representations\n │ (per peer)\n │ │\n │ READ PATH │\n │ ┌─────────────────────────────┐ │\n │ │ generate-context.py (cron) │←───────────┘\n │ │ Token-budgeted files that │\n │ │ survive compaction │\n │ └──────────┬──────────────────┘\n │ │\n │ ↓\n │ HONCHO-CONTEXT.md (shared)\n │ agents/*/HONCHO-CONTEXT.md (per-agent)\n │ │\n │ ↓\n │ Every session loads reasoned context\n │ (main, cron, isolated, sub-agent)\n │\n │ ┌─────────────────────────────┐\n └───→│ query.py (on-demand) │←── Agent mid-conversation\n │ \"What did we decide about │ needs to recall something\n │ pricing last week?\" │\n └─────────────────────────────┘\n\nKey Features\nCompaction-Proof Memory\n\nContext files are workspace files. They survive compaction, session resets, and gateway restarts. Every session starts with reasoned context automatically, not just the last few messages.\n\nCross-Agent Memory\n\nAll agents feed into the same Honcho workspace. Agent A learns something on Monday. Agent B can access that understanding on Tuesday through shared context generation. No manual syncing, no copy-pasting between workspaces.\n\nToken-Budgeted Context\n\nGenerated context files are budget-controlled (~1000-1500 tokens shared, ~500-800 per agent). No prompt bloat. Honcho distills weeks of conversations into dense, relevant context that fits cleanly in the system prompt.\n\nReasoning Levels\n\nOn-demand queries support five reasoning levels: minimal, low, medium, high, max. Quick lookups stay cheap. Deep analysis gets thorough treatment.\n\nHoncho-First, Files as Fallback\n\nDesigned to complement existing file-based memory (MEMORY.md, memory/*.md), not replace it. Honcho handles reasoned recall. Local files handle raw logs and structured data. If Honcho is unavailable, file-based memory still works.\n\nPrerequisites\nPython 3.10+\nHoncho account + API key (app.honcho.dev)\npip install honcho-ai\nSetup\n1. Install SDK\npip install honcho-ai\n\n2. Configure credentials\n\nCreate ~/.config/honcho/credentials.json:\n\n{\n \"api_key\": \"your-honcho-api-key\",\n \"workspace_id\": \"your-workspace-name\"\n}\n\n3. Initialize workspace and peers\npython3 scripts/setup.py --workspace \"my-workspace\" --peers \"user,agent1,agent2\"\n\n\nCreates the workspace and registers every entity (users and agents) that Honcho will track and build representations for.\n\nUsage\nFeed conversations\npython3 scripts/feed.py\n\n\nAuto-discovers active agent session transcripts, attributes messages to correct peers, tracks sync position to avoid duplicates, feeds in batches. Run after sessions or on a cron schedule.\n\nQuery on-demand\n\nWhen an agent needs context it doesn't have:\n\npython3 scripts/query.py \"What was the decision on pricing?\"\npython3 scripts/query.py \"What are the user's priorities right now?\" --peer user-name\npython3 scripts/query.py \"Full analysis of this week's performance\" --level high\n\nGenerate context files\npython3 scripts/generate-context.py # all agents\npython3 scripts/generate-context.py --agent agent-name # specific agent\npython3 scripts/generate-context.py --query \"topic\" # on-demand deep query\n\n\nProduces:\n\nHONCHO-CONTEXT.md in workspace root (shared across all agents)\nHONCHO-CONTEXT.md in each agent's workspace (agent-specific)\nCheck sync status\npython3 scripts/feed.py --status\n\nCron Automation\n\nRun hourly during active hours to keep memory fresh:\n\n0 9-23 * * * cd /workspace && python3 scripts/feed.py && python3 scripts/generate-context.py\n\n\nOpenClaw cron tool config:\n\n{\n \"name\": \"honcho-sync\",\n \"schedule\": {\"kind\": \"cron\", \"expr\": \"0 9-23 * * *\", \"tz\": \"America/Los_Angeles\"},\n \"sessionTarget\": \"isolated\",\n \"payload\": {\n \"kind\": \"agentTurn\",\n \"message\": \"Run: cd /workspace && source .venv/bin/activate && python3 scripts/feed.py && python3 scripts/generate-context.py. Report summary.\",\n \"timeoutSeconds\": 180\n },\n \"delivery\": {\"mode\": \"none\"}\n}\n\nMulti-Agent Setup\n\nFor fleets with multiple agents, each agent gets:\n\nIts own peer in Honcho (tracks what that agent said and learned)\nShared context from the root HONCHO-CONTEXT.md (cross-agent awareness)\nAgent-specific context from its workspace HONCHO-CONTEXT.md (role-specific memory)\n\nThe feed script handles attribution automatically. User messages go to the human peer. Each agent's responses go to that agent's peer. Honcho reasons across all of them.\n\nExample: 6-agent fleet\n\nMain agent, social media agent, content agent, community agent, news scanner, expansion agent\nAll feed into one Honcho workspace\nEach gets tailored context files reflecting their role and recent activity\nThe main agent's context includes a cross-agent brief summarizing what every other agent did\nIntegration with Agent Prompts\n\nAdd to each agent's system prompt or operating instructions:\n\n## Memory\n- At session start: read HONCHO-CONTEXT.md for reasoned context from all conversations\n- If you don't remember something or need context you don't have:\n python3 scripts/query.py \"your question\" --peer user-name\n- Honcho reasons over ALL agent conversations. It likely has the answer.\n- Before saying \"I don't know\" about past decisions, projects, or conversations: query Honcho first.\n\nUse Cases\nCron job needs context: Generated context files give every isolated session full awareness of recent activity, decisions, and priorities without carrying conversation history.\nAgent forgot a decision: On-demand query pulls reasoned context about any topic from weeks of conversations.\nNew agent joins the fleet: Reads shared HONCHO-CONTEXT.md and immediately understands the user, active projects, and current priorities.\nPost-compaction recovery: Context files survive compaction. Agent reloads and continues without losing the thread.\nCross-agent coordination: Social media agent needs to know what the content agent published. Shared context provides this automatically.\nHow Honcho's Reasoning Works\n\nHoncho's data model: Workspaces → Peers → Sessions → Messages. Workspaces isolate environments. Peers are any entity that persists and changes over time (users, agents). Sessions are conversation threads between peers. Messages are units of data that trigger reasoning.\n\nWhen messages are written, Honcho doesn't just store them. Custom reasoning models (Neuromancer) perform formal logical reasoning in the background to generate conclusions about each peer. These conclusions compound into representations that evolve with every new interaction. When you query, you get reasoned context derived from representations, not raw message retrieval.\n\nObservations: Deductive and inductive conclusions drawn from conversations\nRepresentations: Synthesized, evolving understanding of each peer\nContext retrieval: Token-budgeted, combining representations, recent activity, and reasoned observations\n\nFor full details on Honcho's architecture: docs.honcho.dev\n\nCost: ~$2/million tokens ingested. Storage and retrieval free. Most multi-agent setups run $4-6/month.\n\nReferences\nreferences/api-reference.md — Full Honcho SDK API documentation\nreferences/architecture.md — Integration patterns and data flow diagrams\nHoncho docs\nPlastic Labs" }, "trust": { "sourceLabel": "tencent", "provenanceUrl": "https://clawhub.ai/JKILLR/honcho-memory", "publisherUrl": "https://clawhub.ai/JKILLR/honcho-memory", "owner": "JKILLR", "version": "1.0.4", "license": null, "verificationStatus": "Indexed source record" }, "links": { "detailUrl": "https://openagent3.xyz/skills/honcho-memory", "downloadUrl": "https://openagent3.xyz/downloads/honcho-memory", "agentUrl": "https://openagent3.xyz/skills/honcho-memory/agent", "manifestUrl": "https://openagent3.xyz/skills/honcho-memory/agent.json", "briefUrl": "https://openagent3.xyz/skills/honcho-memory/agent.md" } }