nodal-agents 0.6.0 → 0.6.2

package/README.md

@@ -10,6 +10,7 @@ A self-hosted platform for building and orchestrating a **team of AI agents** on
 [![Node](https://img.shields.io/badge/node-22%2B-339933?logo=node.js&logoColor=white)](https://nodejs.org)
 [![TypeScript](https://img.shields.io/badge/typescript-strict-3178c6?logo=typescript&logoColor=white)](https://www.typescriptlang.org)
 [![Docs](https://img.shields.io/badge/docs-online-3565ff)](https://kwintspiracy.github.io/nodal-agents/)
+[![Changelog](https://img.shields.io/badge/changelog-v0.6.2-d4ff2e)](CHANGELOG.md)
 
 | Home dashboard — light theme | Agent detail — dark theme |
 | :---: | :---: |
@@ -60,7 +61,7 @@ Every agent runs on the model **you** choose, with the tools **you** grant, in a
 ### 🤝 Agents that actually finish
 
 - **Orchestrators that finish** — pick the delegation style per request: route to one specialist and resume on its result, or fan work out to many sub-agents in parallel and compile the answer.
-- **Memory that compounds** — persistent facts auto-injected into every job, plus chat-thread continuity (your agent remembers what it said 30 seconds ago — and yesterday).
+- **Memory that compounds** — durable facts auto-injected into every job, ranked by relevance to the task at hand; full-text `search_history` recall over everything the team has ever done; and a background curator that keeps the fact store clean. Plus chat-thread continuity that actually holds a conversation across slow tasks and natural pauses.
 - **Self-improving (opt-in)** — after a substantial job an agent can reflect and write itself a reusable skill; a weekly curator consolidates and prunes them. Every learned skill is reviewable, assignable, and revocable.
 - **Self-extending ROOT** — designate an orchestrator as ROOT and let it create skills, agents, MCP servers, and connectors on your behalf — gated by per-grant toggles and an autonomy level (propose-confirm → fully-autonomous).
 - **Human-in-the-loop** — risky tools pause for approval, and the prompt leads with the agent's plain-language explanation of *what* it wants to do and *why* (plus any ⚠️ impact) — not a wall of raw shell.
@@ -72,11 +73,11 @@ Every agent runs on the model **you** choose, with the tools **you** grant, in a
 **Models** — Anthropic · OpenAI · Google · Groq · Mistral · OpenRouter · **native DeepSeek** (`api.deepseek.com`) · **native MiniMax** (`api.minimax.io`) · any local model (LM Studio, Ollama). One key per provider; each agent picks its own model.
 
 **Connectors** — multi-instance (Gmail perso *and* boulot on one install), managed from the dashboard:
-- *OAuth* — Gmail · Google Drive · Sheets · Docs · Calendar · Notion · Airtable
-- *API key* — Notion · Airtable · Apify · Firecrawl · Tavily
+- *OAuth* — Airtable · Calendar · Docs · Gmail · Google Drive · Notion · Sheets
+- *API key* — Airtable · Apify · Firecrawl · Notion · Tavily
 
 **MCP servers** — over Streamable HTTP *and* stdio, plus add/edit your own:
-Stripe · GitHub · Supabase · Notion · Linear · Sentry · n8n · PostgreSQL · Playwright · Perplexity · Apify · Filesystem · Git · Fetch · Sequential Thinking · Composio · Cogni Cortex
+Apify · Blender · Cogni Cortex · Composio · Fetch · Filesystem · Git · GitHub · KeyShot · Linear · n8n · Notion · Perplexity · Photoshop · Playwright · PostgreSQL · Sentry · Stripe · Supabase · Unity · Unreal Engine
 
 **Skills** — built-in: office editing (Excel/Word/PowerPoint), shell execution, Obsidian, Telegram etiquette, task planning, markdown output, citation discipline, research methodology, results delivery, language mirroring, HTML design. Plus **install any community `SKILL.md`** from GitHub / skills.sh / ClawHub — and agents can **write their own**.
 
@@ -164,7 +165,7 @@ pnpm deps:check
 
 ## Status
 
-Used daily by the maintainer; stable enough for personal production. **Pre-1.0** — breaking changes are still possible between minors.
+Used daily by the maintainer; stable enough for personal production. **Pre-1.0** — breaking changes are still possible between minors. See the [changelog](CHANGELOG.md) for what shipped in each release.
 
 **On the roadmap** (genuine, not vaporware):
 

package/cli.js

@@ -12171,9 +12171,288 @@ var init_community_skill_catalog = __esm({
 });
 
 // ../../packages/shared/src/model-catalog.ts
+var MODEL_CATALOG, VISION_MODEL_IDS;
 var init_model_catalog = __esm({
   "../../packages/shared/src/model-catalog.ts"() {
     "use strict";
+    MODEL_CATALOG = {
+      // ─── Native DeepSeek (api.deepseek.com) ─────────────────────────────────────
+      // Model IDs verified against DeepSeek API docs (June 2026).
+      // Context windows from DeepSeek's documentation: both deepseek-chat and
+      // deepseek-reasoner support 64K output / 128K context.
+      deepseek: [
+        {
+          modelId: "deepseek-chat",
+          label: "DeepSeek Chat (V3)",
+          // deepseek-chat is the production alias for DeepSeek V3. Standard tool
+          // calling with forced tool_choice supported.
+          capabilities: { tools: true, forcedToolChoice: true },
+          contextWindow: 128e3
+        },
+        {
+          modelId: "deepseek-reasoner",
+          label: "DeepSeek Reasoner (R1)",
+          // DeepSeek R1 — a reasoning/thinking model. Requires `thinking: {type: 'enabled'}`
+          // in the request (the deepseek.ts provider sets this automatically when
+          // reasoning:true). Also requires reasoning_content to be echoed back on
+          // assistant messages with tool_calls (the deepseek fetch shim handles this).
+          capabilities: { tools: true, forcedToolChoice: true, reasoning: true },
+          contextWindow: 128e3
+        }
+      ],
+      // ─── Native MiniMax (api.minimax.io/anthropic) ───────────────────────────────
+      // Current MiniMax text models, from the official docs
+      // (platform.minimax.io/docs/api-reference/text-chat — June 2026). The M2/M3
+      // line supersedes M1, which is no longer listed, so it's dropped here. The
+      // full lineup also includes M2.1/M2.5 and `-highspeed` variants — use the
+      // model picker's "Custom…" field for those; this is the curated short list.
+      // forcedToolChoice:false across the M-series — the MiniMax endpoint rejects a
+      // forced tool_choice with a 400/404 (observed on M3 via OpenRouter); the
+      // runtime completion floor covers it. Context windows are docs-sourced where
+      // available, else a conservative 200K until a live probe confirms.
+      minimax: [
+        {
+          modelId: "MiniMax-M3",
+          label: "MiniMax M3",
+          // Newest flagship — a reasoning model (adaptive `thinking`), 1M context.
+          capabilities: { tools: true, forcedToolChoice: false, reasoning: true },
+          contextWindow: 1048576
+        },
+        {
+          modelId: "MiniMax-M2.7",
+          label: "MiniMax M2.7",
+          // Latest M2-series — non-reasoning per MiniMax docs.
+          capabilities: { tools: true, forcedToolChoice: false },
+          contextWindow: 2e5
+        },
+        {
+          modelId: "MiniMax-M2",
+          label: "MiniMax M2",
+          // Stable previous-generation baseline — non-reasoning.
+          capabilities: { tools: true, forcedToolChoice: false },
+          contextWindow: 2e5
+        }
+      ],
+      anthropic: [
+        {
+          modelId: "claude-opus-4-8",
+          label: "Claude Opus 4.8",
+          capabilities: { tools: true, forcedToolChoice: true },
+          contextWindow: 2e5
+        },
+        {
+          modelId: "claude-sonnet-4-6",
+          label: "Claude Sonnet 4.6",
+          capabilities: { tools: true, forcedToolChoice: true },
+          contextWindow: 2e5
+        },
+        {
+          modelId: "claude-haiku-4-5-20251001",
+          label: "Claude Haiku 4.5",
+          capabilities: { tools: true, forcedToolChoice: true },
+          contextWindow: 2e5
+        }
+      ],
+      openai: [
+        {
+          modelId: "gpt-5",
+          label: "GPT-5",
+          capabilities: { tools: true, forcedToolChoice: true },
+          contextWindow: 4e5
+        },
+        {
+          modelId: "gpt-5-mini",
+          label: "GPT-5 mini",
+          capabilities: { tools: true, forcedToolChoice: true },
+          contextWindow: 4e5
+        }
+      ],
+      google: [
+        {
+          modelId: "gemini-2.0-flash",
+          label: "Gemini 2.0 Flash",
+          capabilities: { tools: true, forcedToolChoice: true },
+          contextWindow: 1048576
+        },
+        {
+          modelId: "gemini-2.5-pro",
+          label: "Gemini 2.5 Pro",
+          capabilities: { tools: true, forcedToolChoice: true },
+          contextWindow: 1048576
+        }
+      ],
+      groq: [
+        {
+          modelId: "llama-3.3-70b-versatile",
+          label: "Llama 3.3 70B",
+          capabilities: { tools: true, forcedToolChoice: true },
+          contextWindow: 131072
+        }
+      ],
+      mistral: [
+        {
+          modelId: "mistral-large-latest",
+          label: "Mistral Large",
+          capabilities: { tools: true, forcedToolChoice: true },
+          contextWindow: 131072
+        }
+      ],
+      // OpenRouter models are namespaced by sub-vendor (anthropic/, deepseek/, …).
+      // The UI groups them by that vendor (see modelGroupLabel). Tested + working
+      // routes. `forcedToolChoice` is per-model: most accept tool_choice:'required';
+      // MiniMax M3 does not (some of its OpenRouter endpoints reject the forced
+      // value), so it runs on 'auto' + the runtime floor.
+      openrouter: [
+        // Anthropic
+        {
+          modelId: "anthropic/claude-haiku-4.5",
+          label: "Claude Haiku 4.5",
+          capabilities: { tools: true, forcedToolChoice: true },
+          contextWindow: 2e5
+        },
+        {
+          modelId: "anthropic/claude-opus-4.7",
+          label: "Claude Opus 4.7",
+          capabilities: { tools: true, forcedToolChoice: true },
+          contextWindow: 1e6
+        },
+        {
+          modelId: "anthropic/claude-opus-4.7-fast",
+          label: "Claude Opus 4.7 (fast)",
+          capabilities: { tools: true, forcedToolChoice: true },
+          contextWindow: 1e6
+        },
+        {
+          modelId: "anthropic/claude-opus-4.8",
+          label: "Claude Opus 4.8",
+          capabilities: { tools: true, forcedToolChoice: true },
+          contextWindow: 1e6
+        },
+        {
+          modelId: "anthropic/claude-opus-4.8-fast",
+          label: "Claude Opus 4.8 (fast)",
+          capabilities: { tools: true, forcedToolChoice: true },
+          contextWindow: 1e6
+        },
+        {
+          modelId: "anthropic/claude-sonnet-4.6",
+          label: "Claude Sonnet 4.6",
+          capabilities: { tools: true, forcedToolChoice: true },
+          contextWindow: 1e6
+        },
+        // DeepSeek
+        {
+          modelId: "deepseek/deepseek-v3.2",
+          label: "DeepSeek V3.2",
+          capabilities: { tools: true, forcedToolChoice: true },
+          contextWindow: 131072
+        },
+        {
+          modelId: "deepseek/deepseek-v4-flash",
+          label: "DeepSeek V4 Flash",
+          // DeepSeek V4 is a reasoning model (reasoning_content / reasoning effort).
+          // Flagging reasoning makes the OpenRouter provider enable reasoning + round-trip
+          // its reasoning_details across tool calls — without it the model misbehaves
+          // (gathers then emits no answer). Matches how Hermes drives DeepSeek.
+          capabilities: { tools: true, forcedToolChoice: true, reasoning: true },
+          contextWindow: 1048576,
+          providerOrder: ["deepseek"]
+        },
+        {
+          modelId: "deepseek/deepseek-v4-pro",
+          label: "DeepSeek V4 Pro",
+          capabilities: { tools: true, forcedToolChoice: true, reasoning: true },
+          contextWindow: 1048576,
+          providerOrder: ["deepseek"]
+        },
+        // Google
+        {
+          modelId: "google/gemini-3.1-flash-lite-preview",
+          label: "Gemini 3.1 Flash Lite (preview)",
+          capabilities: { tools: true, forcedToolChoice: true },
+          contextWindow: 1048576
+        },
+        {
+          modelId: "google/gemini-3.1-pro-preview",
+          label: "Gemini 3.1 Pro (preview)",
+          capabilities: { tools: true, forcedToolChoice: true },
+          contextWindow: 1048576
+        },
+        {
+          modelId: "google/gemini-3.5-flash",
+          label: "Gemini 3.5 Flash",
+          capabilities: { tools: true, forcedToolChoice: true },
+          contextWindow: 1048576
+        },
+        {
+          modelId: "google/gemma-4-31b-it",
+          label: "Gemma 4 31B-IT",
+          capabilities: { tools: true, forcedToolChoice: true },
+          contextWindow: 262144
+        },
+        // MiniMax
+        {
+          modelId: "minimax/minimax-m3",
+          label: "MiniMax M3",
+          // A reasoning model. Some of its OpenRouter endpoints reject a FORCED
+          // tool_choice ('required') → we send 'auto' (forcedToolChoice:false).
+          // reasoning:true makes the provider return reasoning_details so the runner
+          // can round-trip them across tool-call turns.
+          capabilities: { tools: true, forcedToolChoice: false, reasoning: true },
+          contextWindow: 1048576
+        },
+        // Z.ai (GLM)
+        {
+          modelId: "z-ai/glm-5.2",
+          label: "GLM 5.2",
+          // A reasoning model (reasoning effort high/xhigh), strong at coding + tool
+          // use. reasoning:true → provider returns reasoning_details for round-trip.
+          // forcedToolChoice:false: send 'auto' (don't risk a rejected 'required' on
+          // a reasoning model; workers are 'auto' after turn 1 anyway).
+          capabilities: { tools: true, forcedToolChoice: false, reasoning: true },
+          contextWindow: 1048576
+        },
+        // Moonshot (Kimi)
+        {
+          modelId: "moonshotai/kimi-k2.7-code",
+          label: "Kimi K2.7 Code",
+          // Always operates in thinking mode and preserves reasoning_content across
+          // turns (like DeepSeek) → reasoning:true is required so the provider
+          // round-trips reasoning_details. forcedToolChoice:false for the same reason
+          // as the other reasoning models.
+          capabilities: { tools: true, forcedToolChoice: false, reasoning: true },
+          contextWindow: 262144
+        }
+      ]
+    };
+    VISION_MODEL_IDS = /* @__PURE__ */ new Set([
+      // OpenRouter (verified via /api/v1/models)
+      "anthropic/claude-haiku-4.5",
+      "anthropic/claude-opus-4.7",
+      "anthropic/claude-opus-4.7-fast",
+      "anthropic/claude-opus-4.8",
+      "anthropic/claude-opus-4.8-fast",
+      "anthropic/claude-sonnet-4.6",
+      "google/gemini-3.1-flash-lite-preview",
+      "google/gemini-3.1-pro-preview",
+      "google/gemini-3.5-flash",
+      "google/gemma-4-31b-it",
+      "minimax/minimax-m3",
+      "moonshotai/kimi-k2.7-code",
+      // Native-provider forms
+      "claude-opus-4-8",
+      "claude-sonnet-4-6",
+      "claude-haiku-4-5-20251001",
+      "gpt-5",
+      "gpt-5-mini",
+      "gemini-2.0-flash",
+      "gemini-2.5-pro",
+      "mistral-large-latest",
+      "MiniMax-M3"
+    ]);
+    for (const entries of Object.values(MODEL_CATALOG)) {
+      for (const e of entries) e.capabilities.vision = VISION_MODEL_IDS.has(e.modelId);
+    }
   }
 });
 
@@ -12516,6 +12795,14 @@ var init_jobs = __esm({
         chatId: text("chat_id"),
         systemPrompt: text("system_prompt"),
         messages: jsonb("messages").default(sql`'[]'::jsonb`),
+        /**
+         * Flattened plain-text transcript (task + assistant text + tool outputs +
+         * result) for full-text episodic search. Populated at job completion by
+         * flattenTranscript(). A generated `search_tsv tsvector` column + GIN index
+         * (raw SQL, migration 0050 — not expressible in the Drizzle schema builder)
+         * makes it queryable by the `search_history` builtin.
+         */
+        searchText: text("search_text"),
         toolsUsed: text("tools_used").array().default(sql`'{}'::text[]`),
         turn: integer("turn").default(0),
         result: text("result"),

package/migrations/0050_jobs_search_fts.sql

@@ -0,0 +1,4 @@
+ALTER TABLE "agent_jobs" ADD COLUMN "search_text" text;--> statement-breakpoint
+ALTER TABLE "agent_jobs" ADD COLUMN "search_tsv" tsvector GENERATED ALWAYS AS (to_tsvector('simple', coalesce("search_text", ''))) STORED;--> statement-breakpoint
+CREATE INDEX "idx_agent_jobs_search_tsv" ON "agent_jobs" USING gin ("search_tsv");--> statement-breakpoint
+UPDATE "agent_jobs" SET "search_text" = trim(coalesce("task", '') || ' ' || coalesce("result", '')) WHERE "search_text" IS NULL AND (coalesce("task", '') <> '' OR coalesce("result", '') <> '');

package/migrations/meta/_journal.json

@@ -344,6 +344,13 @@
       "when": 1782900000000,
       "tag": "0049_skill_assignment_files_writable",
       "breakpoints": true
+    },
+    {
+      "idx": 50,
+      "version": "7",
+      "when": 1783000000000,
+      "tag": "0050_jobs_search_fts",
+      "breakpoints": true
     }
   ]
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "nodal-agents",
-  "version": "0.6.0",
+  "version": "0.6.2",
   "description": "Local-first AI agent platform with a web dashboard — install in one command.",
   "license": "MIT",
   "repository": {