abtars 0.2.1-alpha.10 → 0.2.1-alpha.11

package/README.md

@@ -1,29 +1,36 @@
 # abTARS
 
-A personal AI agent that runs on your machine, talks to you via Telegram/Discord/IRC, and remembers everything. Use frontier models through existing subscriptions — no per-token billing.
+**Autonomous AI agent with persistent memory, self-healing, and peer-to-peer communication. Your hardware, your rules.**
 
-## What it does
+abTARS runs 24/7 on your machine — talks to you via Telegram/Discord/IRC, remembers everything across sessions, executes scheduled tasks, and recovers from failures without intervention. Use frontier models through existing CLI subscriptions at zero marginal cost.
 
-- **Always-on agent** — runs 24/7, responds on Telegram/Discord, executes tasks on schedule
-- **Persistent memory** — remembers conversations, learns preferences, extracts facts overnight
-- **Multi-model** — Claude, Gemini, OpenRouter (40+ models), ollama (local). Switch by changing config
-- **Self-healing** — 4-layer watchdog, auto model fallback, survives crashes and network drops
-- **Extensible** — skills system, MCP integration, agent-to-agent communication
+## Why abTARS
+
+- 🧠 **Memory that curates itself** — multi-layer recall (5 search stages + reranking), nightly sleep maintenance, emotion tracking, contradiction detection, Memory Darwinism
+- 🛡️ **Self-hosted, defense-in-depth** — classified memory (4 tiers, encrypted at rest), role-based access, injection scanning, secrets vault
+- 🔄 **Runs months unattended** — 5-layer supervision, leaky-bucket model fallback, self-healing agent, standby-aware recovery
+- 🤝 **Agent-to-agent** — P2P communication with Ed25519 signatures, mDNS wake-up, IRC coordination channels
+- 💰 **Zero idle cost** — no LLM calls at rest, CLI subscription parasitism, budget sleep tiers
+
+→ **[Why abTARS vs OpenClaw & Hermes](https://aksika.github.io/abtars/why)**
+
+## Architecture
 
 ```
-You (Telegram/Discord/IRC)
+You (Telegram / Discord / IRC / API client)
   │
   ▼
-abTARS (agent brain)
-  ├── Memory (abmind — persistent, searchable, encrypted)
-  ├── Personality (customizable identity + learned behavior)
-  ├── Tools (browse, recall, store, bash, MCP servers)
-  ├── Skills (hot-reloadable, self-authoring)
+abTARS (bridge)
+  ├── abmind (memory — in-process, multi-layer recall, encrypted)
+  ├── Skills (core + self-authored during sleep + downloadable)
+  ├── Tools (browse, bash, MCP, peer_ask)
+  ├── Tasks (cron scheduler + retry + DoD checks)
+  ├── Agent Swarm (async background sessions)
   │
-  ├── kiro-cli     → Claude (AWS subscription)
-  ├── gemini-cli   → Gemini
-  ├── Codex / Claude Code → OpenAI / Anthropic
-  └── Direct API   → ollama, OpenRouter, any OpenAI-compatible
+  ├── kiro-cli        → Claude, DeepSeek, MiniMax, Qwen (free tier)
+  ├── gemini-cli      → Gemini 2.5
+  ├── Direct API      → ollama, OpenRouter, any OpenAI-compatible
+  └── Peers           → other abTARS instances via /v1/chat/completions
 ```
 
 ## Quick Start
@@ -37,37 +44,84 @@ abtars onboard
 sudo $(which abtars) daemon install
 ```
 
-Full installation guide: **[docs/wiki/install.md](docs/wiki/install.md)**
+After install, configure `~/.abtars/config/.env` with your Telegram bot token and at least one model provider. Full guide: **[docs/wiki/install.md](docs/wiki/install.md)**
 
-## Documentation
+## Features
 
-Docs live in the repo at **[docs/wiki/](docs/wiki/)** — always up to date with the code.
+### Memory (abmind)
 
-- [Installation](docs/wiki/install.md)
-- [Configuration](docs/wiki/commands.md)
-- [Commands](docs/wiki/cli.md)
-- [Memory System (abmind)](https://github.com/aksika/abmind)
-- [Skills & Extensions](docs/wiki/skills.md)
-- [Deployment & Supervision](docs/wiki/supervision.md)
+Standalone package — works with abTARS, Kiro CLI, Gemini CLI, Claude Code, Codex, Hermes, or any MCP client.
+
+- Multi-layer recall: 5 search stages (porter FTS5 → trigram → binary signatures → vector embeddings → entity graph) + 7 post-processing layers (cross-stage penalty, context boost, emotion boost, spacing boost, quality boost, MMR reranking, interference detection)
+- Agglutinative language support (Hungarian, Finnish, Turkish) — QWERTZ fallback, substring windows
+- 25 emotion types with per-memory scoring, emotional arcs per topic
+- 12-step nightly sleep: extract, consolidate, prune, detect contradictions, fix translations
+- NATO-inspired classification: confidentiality × trust × integrity × credibility
+- AES-256-GCM encrypted secrets vault with auto-redaction from history
+- Memory Darwinism — unused memories fade, recalled memories strengthen
+
+### Reliability
+
+- **L1** Heartbeat — standby detection, bridge.lock, task dispatch
+- **L2** In-process watchdog — detects stuck event loops
+- **L3** External watchdog — catches dead PIDs, stale heartbeats. Circuit breaker prevents restart storms
+- **L4** OS supervisor — launchd (macOS) / systemd (Linux) restarts the watchdog itself
+- **L5** Preventive daily restart — eliminates memory leaks
+- **Model health** — leaky-bucket per model, progressive penalties, arbitrarily long fallback chains
+- **Self-healing agent** — diagnoses failed tasks, attempts repair, suspends after 3 failures
+
+### Agent Swarm
+
+Main agent spawns independent background sessions (own context, own tool loop). Results auto-inject into the parent's next prompt. Up to 3 concurrent. `/wait` injects instructions into running sessions.
+
+### Peer-to-Peer
 
-> **Note:** The [aksika.github.io](https://aksika.github.io/abtars/) site may be outdated. Always refer to the docs in this repo.
+Multiple abTARS instances communicate via OpenAI-compatible `/v1/chat/completions` endpoint. Bearer auth per peer, Ed25519 signatures, mDNS wake-up for firewalled peers, hop-limit loop prevention.
+
+### Security
+
+- Platform-level access: only registered chatId/userId can reach the agent
+- Role-based: master/user/guest — commands, tools, memory all gated
+- Secrets vault: AES-256-GCM, scrypt-derived key, auto-encrypt on ingest
+- Injection scanner on all inbound messages
+- SSRF guard on browser agent
+- Credential redaction in all logs and exports
 
 ## Supported Transports
 
-| Transport | Provider | 
-|-----------|----------|
-| ACP | kiro-cli, gemini-cli, Codex, Claude Code |
-| Direct API | OpenRouter, ollama, any OpenAI-compatible endpoint |
-| Hooks | abmind hooks (standalone CLI agents) |
+| Transport | Providers |
+|-----------|-----------|
+| ACP (recommended) | kiro-cli, gemini-cli |
+| Direct API | ollama, OpenRouter, any OpenAI-compatible endpoint |
+| Hooks (standalone) | abmind lifecycle hooks on any CLI agent |
 
 ## Requirements
 
 - Node.js 22+
-- A Telegram bot token
-- At least one model provider (kiro-cli, gemini-cli, ollama, OpenRouter, Codex, or Claude Code)
+- A Telegram bot token (Discord/IRC optional)
+- At least one model provider
 
 Optional: ollama + `nomic-embed-text` for memory embeddings.
 
+## Documentation
+
+- [Installation](docs/wiki/install.md)
+- [Configuration](docs/wiki/commands.md)
+- [CLI Reference](docs/wiki/cli.md)
+- [Memory System (abmind)](https://github.com/aksika/abmind)
+- [Skills & Extensions](docs/wiki/skills.md)
+- [Deployment & Supervision](docs/wiki/supervision.md)
+
+Full docs: **[aksika.github.io/abtars](https://aksika.github.io/abtars/)**
+
+## Numbers
+
+- 1794 tests (abtars 1016 + abmind 778)
+- 5 agent types (professor, dreamy, browsie, coding, cron)
+- 5-layer supervision stack
+- 3 platform adapters + OpenAI-compatible API
+- 12-step nightly memory maintenance
+
 ## Development
 
 ```bash
@@ -76,6 +130,11 @@ cd abtars && npm install && npm run build
 npm test
 ```
 
+## Community
+
+- **Discord:** [Join](https://discord.gg/pj2qbWJT8)
+- **GitHub:** [aksika/abtars](https://github.com/aksika/abtars) · [aksika/abmind](https://github.com/aksika/abmind)
+
 ## License
 
 Apache-2.0

package/bundle/{_registry.generated-N74SMTUJ.js → _registry.generated-AJRCDWPQ.js}

@@ -1,26 +1,27 @@
 #!/usr/bin/env node
  import { createRequire as __bundleCreateRequire } from 'node:module'; import { fileURLToPath as __bundleFileURLToPath } from 'node:url'; import { dirname as __bundleDirname } from 'node:path'; const require = __bundleCreateRequire(import.meta.url); const __chunk_filename = __bundleFileURLToPath(import.meta.url); const __chunk_dirname = __bundleDirname(__chunk_filename);
-import {
-  sleep_exports
-} from "./chunk-7I5C2U3C.js";
 import {
   browser_exports
-} from "./chunk-VYR25H37.js";
-import "./chunk-LDKSCXGL.js";
+} from "./chunk-UBQJA7N4.js";
+import "./chunk-4IQ2HVXR.js";
 import {
   hotskills_exports
 } from "./chunk-RE3F3CFW.js";
-import "./chunk-HEHD3GG5.js";
+import {
+  sleep_exports
+} from "./chunk-DWCCYHRT.js";
+import "./chunk-UUOUTFBM.js";
 import "./chunk-JAJ3DUQ2.js";
 import "./chunk-BYDUMHXT.js";
 import "./chunk-UDZIZB5F.js";
 import "./chunk-TBLYGCPQ.js";
 import "./chunk-GPL57SRN.js";
 import "./chunk-MZWMYN4O.js";
-import "./chunk-KSEIWT4T.js";
-import "./chunk-EX2SRTUE.js";
+import "./chunk-ZAUTS7MH.js";
 import "./chunk-PZE3J7ER.js";
+import "./chunk-EX2SRTUE.js";
 import "./chunk-2BY6I4P5.js";
+import "./chunk-DGS7J4P6.js";
 import "./chunk-MJ6PHMOK.js";
 import "./chunk-7K2YZTLD.js";
 
@@ -33,4 +34,4 @@ var capabilities = [
 export {
   capabilities
 };
-//# sourceMappingURL=_registry.generated-N74SMTUJ.js.map
+//# sourceMappingURL=_registry.generated-AJRCDWPQ.js.map

package/bundle/{_registry.generated-G2FSGHNY.js.map → _registry.generated-AJRCDWPQ.js.map}

@@ -2,6 +2,6 @@
   "version": 3,
   "sources": ["../src/capabilities/_registry.generated.ts"],
   "sourcesContent": ["// AUTO-GENERATED by esbuild.config.js \u2014 do not edit\n// Static capability registry for bundled builds\n\nimport * as cap0 from \"./browser/index.js\";\nimport * as cap1 from \"./hotskills/index.js\";\nimport * as cap2 from \"./sleep/index.js\";\n\nexport const capabilities = [\n  { name: \"browser\", module: cap0 },\n  { name: \"hotskills\", module: cap1 },\n  { name: \"sleep\", module: cap2 },\n];\n"],
-  "mappings": ";;;;;;;;;;;;;;;;;;;;;;;;;;;AAOO,IAAM,eAAe;AAAA,EAC1B,EAAE,MAAM,WAAW,QAAQ,gBAAK;AAAA,EAChC,EAAE,MAAM,aAAa,QAAQ,kBAAK;AAAA,EAClC,EAAE,MAAM,SAAS,QAAQ,cAAK;AAChC;",
+  "mappings": ";;;;;;;;;;;;;;;;;;;;;;;;;;;;AAOO,IAAM,eAAe;AAAA,EAC1B,EAAE,MAAM,WAAW,QAAQ,gBAAK;AAAA,EAChC,EAAE,MAAM,aAAa,QAAQ,kBAAK;AAAA,EAClC,EAAE,MAAM,SAAS,QAAQ,cAAK;AAChC;",
   "names": []
 }

package/bundle/abtars-browser.js

@@ -5,9 +5,9 @@ import {
   BrowserTool,
   DomainAllowlist,
   getDefaultSocketPath
-} from "./chunk-KSEIWT4T.js";
-import "./chunk-EX2SRTUE.js";
+} from "./chunk-ZAUTS7MH.js";
 import "./chunk-PZE3J7ER.js";
+import "./chunk-EX2SRTUE.js";
 import "./chunk-2BY6I4P5.js";
 import "./chunk-MJ6PHMOK.js";
 import "./chunk-7K2YZTLD.js";

package/bundle/abtars-cli.js

@@ -2,10 +2,10 @@
  import { createRequire as __bundleCreateRequire } from 'node:module'; import { fileURLToPath as __bundleFileURLToPath } from 'node:url'; import { dirname as __bundleDirname } from 'node:path'; const require = __bundleCreateRequire(import.meta.url); const __chunk_filename = __bundleFileURLToPath(import.meta.url); const __chunk_dirname = __bundleDirname(__chunk_filename);
 import {
   install
-} from "./chunk-VREI2GUO.js";
+} from "./chunk-4NZGCLOX.js";
 import {
   rollback
-} from "./chunk-EZZBO2QK.js";
+} from "./chunk-2SH2U57T.js";
 import {
   acquireLock,
   atomicSwap,
@@ -19,7 +19,7 @@ import {
   readSentinel,
   writeManifest,
   writeSentinel
-} from "./chunk-JRK3UFBS.js";
+} from "./chunk-WMWI3SJ7.js";
 import {
   restart
 } from "./chunk-URAQLQ2U.js";
@@ -314,10 +314,14 @@ var ABTARS_EXCLUDE = [
   "current",
   "bin",
   "app",
+  "app.prev*",
+  "app.staging",
+  "app.broken",
   "logs",
   "node_modules",
   "overflow",
   "browser-socket",
+  "src",
   "bridge.lock",
   "watchdog.lock",
   "browse-spawn.sock",
@@ -578,7 +582,7 @@ async function runInteractive(existing) {
   const { intro, outro, text, select, confirm: confirm2, isCancel, cancel } = await import("./dist-J3T4XVKX.js");
   intro("abtars onboard \u2014 first-time setup");
   const noteEmpty = "press Enter to skip";
-  const { packagePaths: pp, readManifest: rm3 } = await import("./deploy-lib-import-PJXUADPS.js");
+  const { packagePaths: pp, readManifest: rm3 } = await import("./deploy-lib-import-6VJTYXEG.js");
   const mfPaths = pp("abtars");
   const mf = await rm3(mfPaths.manifest);
   const installMode = mf?.installMode ?? "supervised";
@@ -1014,7 +1018,7 @@ async function onboard(opts) {
   }
   if (secrets.length > 0) process.stdout.write(`\u2713 ${secrets.length} secrets \u2192 ${secretDirPath}
 `);
-  const { writeManifest: writeManifest2 } = await import("./deploy-lib-import-PJXUADPS.js");
+  const { writeManifest: writeManifest2 } = await import("./deploy-lib-import-6VJTYXEG.js");
   const mf = await readManifest(paths.manifest);
   if (mf) await writeManifest2(paths.manifest, { ...mf, installMode: answers.installMode });
   process.stdout.write(`\u2713 install mode: ${answers.installMode}
@@ -1600,7 +1604,7 @@ Use --source local (default) or --source npm.
       previousVersion: prior?.version ?? null,
       previousCommit: prior?.commit ?? null,
       installMode: prior?.installMode ?? "supervised",
-      repoRoot: opts.fromLocal && existsSync8(join10(repoRoot, ".git")) ? repoRoot : prior?.repoRoot ?? null
+      repoRoot
     });
     process.stdout.write(`\u2713 manifest updated
 `);
@@ -1695,7 +1699,7 @@ async function postSwapHousekeeping(paths, repoRoot, _staged) {
   }
   process.stdout.write(`\u2713 scripts refreshed (${scriptFiles.length} files)
 `);
-  const { writeWrapper } = await import("./install-6H3APAIJ.js");
+  const { writeWrapper } = await import("./install-EBD7J27R.js");
   await mkdir4(paths.bin, { recursive: true });
   for (const name of installManifest.cliWrappers) {
     await writeWrapper(paths.bin, name, paths.app, false);