# ElizaOS Developer Update (2026-03-05 → 2026-03-11)

> Sources: developer Discord discussions (💬-discussion, 💬-coders, xfn-framework) and linked repos/resources. No GitHub PR merge log was provided for this week; items below reflect what was discussed as in-flight or announced.

---

## 1) Core Framework (Architecture / Plugin System / Agent Runtime)

### v2.0.0 architecture work: “Prompt Batching” scheduler (in progress)
Work on the `v2.0.0` branch is consolidating LLM invocation patterns across the ecosystem into a single subsystem called **prompt batching**. After reviewing ~50–60 plugins, the goal is to unify:

- **Init LLM queries** (startup / boot prompts)
- **Autonomous LLM calls** (background loops, timed triggers, agent autonomy)
- **Evaluator calls** (scoring, validation, guardrails)

…into **one configurable scheduler**, tuned differently for:
- **Frontier models** (optimize throughput / latency / cost with batching)
- **Local models** (optimize queueing, concurrency, and CPU/GPU utilization)

This is described as building on existing **dynamicPromptExecution**, with core concepts already present in the “3.x autonomous system” mentioned in discussion.

**Developer impact**
- Expect fewer ad-hoc “call the model now” paths inside plugins; instead, plugins will enqueue work into a common runtime scheduler.
- This is a prerequisite for more predictable cost controls and consistent concurrency limits across plugins.

**Reference (discussion context):**
- xfn-framework Discord thread (prompt batching + scheduler consolidation)

### Serverless-oriented runtime concepts (exploration)
Framework discussions also surfaced several runtime-level concepts to support cloud/serverless deployment:

- **Lazy loading services**: defer service initialization until first use (reduce cold-start and memory footprint)
- **Outsourcing service work**: push heavy work to external systems/jobs where appropriate
- **In-memory persistence**: reduce “rebuild state on every boot” costs for ephemeral runtimes

Stan noted interest in integrating lazy-loading service patterns into the monorepo structure (work not yet landed).

---

## 2) New Features (Capabilities + Examples)

### Autonomous Economy Protocol (AEP) plugin (announced)
A new plugin integration for **Autonomous Economy Protocol (AEP)** was announced, enabling agents to:
- Earn on-chain payments (AGT token)
- Build a **permanent reputation score (0–10,000)** stored on Basescan
- Participate in a marketplace for agent work
- Earn **1% passive referral income** “in perpetuity”
- Potentially access credit based on reputation (no collateral requirement per announcement)

**Chain:** Base  
**SDK:** `autonomous-economy-sdk` (npm)  
**Eliza integration path mentioned:** `integrations/eliza-plugin/`  
**Actions exposed (TypeScript):**
- `REGISTER_AGENT`
- `BROWSE_MARKET`
- `PROPOSE_DEAL`
- `CHECK_REPUTATION`
- `GET_SEASON1_INFO`

**Season 1 Genesis Program:** 50,000,000 AGT pool through ~May 2026 (points-based claim system).

#### Example: wiring AEP actions into an Eliza agent (conceptual)
Exact ElizaOS plugin registration APIs vary by version; the snippet below focuses on the **action invocation shape** that was described.

```ts
import { AEPClient } from "autonomous-economy-sdk";

const aep = new AEPClient({
  // network: "base",
  // wallet: ...
});

async function registerAgent(metadata: { name: string; description: string }) {
  return aep.runAction("REGISTER_AGENT", {
    name: metadata.name,
    description: metadata.description,
  });
}

async function browseMarket(params: { tags?: string[]; limit?: number }) {
  return aep.runAction("BROWSE_MARKET", params);
}

async function proposeDeal(input: {
  listingId: string;
  terms: string;
  priceAgt: string;
}) {
  return aep.runAction("PROPOSE_DEAL", input);
}

async function checkReputation(address: string) {
  return aep.runAction("CHECK_REPUTATION", { address });
}

async function season1Info() {
  return aep.runAction("GET_SEASON1_INFO", {});
}
```

**Docs/links:**
- npm package (as announced): `autonomous-economy-sdk`  
- AEP integration folder noted in announcement: `integrations/eliza-plugin/`  
- (No GitHub URL was provided in the source data for AEP; if one exists, link it in your internal docs.)

### ZARQ risk intelligence plugin (published)
An ElizaOS plugin was published providing **pre-trade risk scoring** covering **205 tokens** via **ZARQ** crypto risk intelligence infrastructure. This is intended to give trading/DeFi agents a standardized risk signal before execution.

**Developer impact**
- If you maintain trading agents, consider inserting the ZARQ scoring call as an evaluator/guard step prior to placing orders.

*(No repository link was included in the provided data; add it to your plugin registry/docs once available.)*

### Communication automation: elizaOS.news daily video briefings
To address update cadence, the team built **elizaOS.news** with an **automated video briefing workflow** that generates daily summaries.

- Site: https://elizaos.news

While not a framework feature, it affects developer communications: changelogs and operational status updates are expected to become more regular and machine-assisted.

---

## 3) Bug Fixes (Critical Fixes + Technical Context)

### Core: `develop` branch is currently broken (known issue, not yet fixed)
Odilitime noted ongoing work on “the next version” of ElizaOS and explicitly stated the **`develop` branch is currently broken**. No specific stack trace or regression range was provided, but the implication is that the v2.0.0 work will “unblock planned tasks” once stabilized.

**Mitigation for developers**
- Pin to the latest known-good release tag (or a stable commit on main) for production deployments.
- If you must track `develop`, gate CI with integration tests around agent boot + plugin load + at least one autonomous cycle.

### Milady: Neon DB config discovered; permissions/capabilities unresolved
BinaryCookies identified where to configure a **Neon database** for Milady: **under `env` in the JSON config**. However, issues remain around **system permissions/capabilities** (unresolved in the thread).

**Developer note**
- If deploying Milady with managed Postgres (Neon), verify:
  - connection string presence in the JSON `env`
  - runtime permissions (container capabilities / host permissions) required by Milady services you enable

### Milady packaging: APT repository distribution PRs (in progress)
Work was discussed to add **APT repository distribution** (Debian/Ubuntu) to simplify installs. This is a delivery/ops improvement; specific PR numbers were not provided in the dataset.

**Related repo/issue context**
- Milady repo: https://github.com/milady-ai/milady
- Issue referenced in PR activity: https://github.com/milady-ai/milady/issues/71

---

## 4) API Changes (Developer-Facing)

No concrete merged API diffs were provided this week, but the **prompt batching** effort strongly implies upcoming API/behavior changes in the agent runtime:

### Expected direction (based on discussion)
- A shift from direct model calls scattered across plugins to a **central scheduler/queue**
- Potential new configuration surface for:
  - batching strategy
  - concurrency limits
  - model tier selection (frontier vs local)
  - prioritization across init/autonomous/evaluator workloads

**Action for plugin maintainers**
- Audit plugins that perform:
  - model calls during init
  - autonomous loops / interval triggers
  - evaluator/validation calls
- Prepare to migrate those calls to the unified scheduler once the v2 interface is published.

---

## 5) Social Media Integrations (Twitter / Telegram / Discord / Farcaster)

### Twitter/X
- Odilitime indicated that completing the next ElizaOS version will unblock **improved Twitter posts** for `$degenai` and `$elizaos`.

### Discord (timed interactions)
A recurring developer question: “How do I schedule agent-to-agent conversations in Discord similar to `TWITTER_POST_INTERVAL_MIN/MAX`?”

A community answer pointed to:
- Autonomous TypeScript examples: https://github.com/elizaOS/examples/tree/main/autonomous/typescript
- Milady trigger systems: https://github.com/milady-ai/milady

**Developer takeaway**
- For now, implement timed Discord behaviors using the autonomous/trigger patterns shown in those examples rather than expecting a dedicated “Discord interval” config knob.

*(No Telegram/Farcaster plugin changes were mentioned in the provided data.)*

---

## 6) Model Provider Updates (OpenAI / Anthropic / DeepSeek / Voice, etc.)

No provider SDK updates were reported, but two relevant runtime/provider themes emerged:

### Frontier vs local optimization (scheduler-level)
Prompt batching is explicitly designed to tune execution for **frontier** and **local** models, which may affect:
- request batching heuristics
- rate limiting
- evaluator placement (pre/post execution)
- cost and latency behavior

### Voice provider cost pressure (ElevenLabs) + request for Google voice
Developers raised concerns about **ElevenLabs** cost and asked for a working **Google voice** plugin as an alternative. No implementation/PR was referenced yet.

---

## 7) Breaking Changes (V1 → V2 Migration Warnings)

### Warning: v2.0.0 runtime refactor will likely break assumptions in plugins
If your plugin currently:
- calls the LLM directly during initialization
- runs autonomous loops with its own timers and queues
- uses separate evaluator calls outside a shared runtime

…you should anticipate **migration work** when prompt batching lands.

**Recommended prep**
- Encapsulate model calls behind a local abstraction in your plugin so you can swap to the scheduler later.
- Avoid relying on “init-time ordering” of model calls across plugins; a scheduler may reorder or parallelize.
- Add tests that validate:
  - agent boots without model calls completing synchronously
  - autonomous tasks can be deferred
  - evaluator calls can be queued and retried

### Warning: `develop` instability
Given `develop` is currently broken (per discussion), do not treat it as a stable integration base for production. Pin versions and watch for a v2 stabilization announcement.

---

## Links & References
- elizaOS.news (daily automated briefings): https://elizaos.news  
- Autonomous agent examples (TypeScript): https://github.com/elizaOS/examples/tree/main/autonomous/typescript  
- Milady repository: https://github.com/milady-ai/milady  
- Milady issue referenced: https://github.com/milady-ai/milady/issues/71