Integrating Claude Code Native Features to Amplify Instar

[EchoOfDawn]

Context

Anthropic rolled out a suite of features in Jan-Mar 2026 that significantly expand Claude Code's autonomous and remote capabilities: Dispatch, Computer Use, Channels, Remote Control, Scheduled Tasks, Auto Mode, and Claude Code on the Web.

Since Instar is built ON Claude Code (wrapping real CLI processes, not API calls), these features represent a major opportunity — not competition, but amplification. Instar can absorb the infrastructure Anthropic is building and focus even more deeply on what makes it unique.

---

What Anthropic Released

Feature	What It Does
Dispatch	Text Claude from phone → runs tasks on your Mac. Routes dev tasks to Code, research to Cowork.
Computer Use	Claude controls your screen — clicks, types, drags. Tiered app permissions. macOS research preview.
Remote Control	claude --remote-control → continue session from phone/browser. Survives network drops.
Channels	Telegram, Discord, iMessage push events INTO Claude Code sessions. Two-way.
Scheduled Tasks	Cloud tasks (1hr min), Desktop tasks (1min min), /loop for session-scoped polling.
Auto Mode	All actions execute with background LLM safety checks — no permission prompts.
Claude Code on the Web	Cloud VMs — no local machine needed.

---

Integration Opportunities for Instar

1. Channels → Messaging Transport Layer

Instar has its own Telegram integration. Channels could replace the custom bridge, gaining Discord and iMessage support for free while letting Instar focus on orchestration (identity, memory, scheduling) rather than messaging plumbing.

2. Remote Control → Live Session Monitoring

Instar sessions are currently fire-and-forget. Exposing --remote-control on Instar-spawned sessions would let users monitor and course-correct long-running agents from their phone — a "supervised autonomy" mode.

3. Auto Mode → Simplified Permission Handling

Instar currently manages permissions through custom safety gates and tiered trust. Auto Mode could handle routine Tier 0-1 actions, while Instar's gates remain for high-risk Tier 2-3 operations. This would dramatically simplify the permission management code.

4. Computer Use → GUI-Capable Agents

Instar agents are currently CLI-only. With Computer Use, scheduled jobs could check dashboards, take screenshots, manage browser-based tools without APIs, and run visual testing workflows.

5. Desktop Scheduled Tasks → Complementary Scheduling

Claude Code's native cron has a 3-day expiry — Instar's doesn't. But Desktop tasks handle catch-up logic natively and could complement Instar's scheduler for simpler recurring tasks.

6. Claude Code on the Web → Hardware-Free Instar

This is the biggest potential unlock. If Instar sessions could run on Anthropic's cloud VMs, agents run without any dedicated hardware. Major adoption barrier removed. Challenge: state persistence between runs would need a sync strategy.

---

What Instar Still Uniquely Provides

Even with all these native features, Claude Code still lacks:

• Persistent identity across sessions (structural grounding, Self-Knowledge Tree)
• Accumulated memory that survives session boundaries (JSONL + FTS5 + Mem0)
• Self-evolution — agents that improve their own capabilities over time
• Multi-agent orchestration — multiple agents sharing state
• Decision accountability — full audit trail with reasoning (DecisionJournal)
• Relationship continuity — tracking contacts and conversations across sessions
• Production-grade observability — telemetry, anomaly detection, cost tracking

The strategic frame: Anthropic builds the plumbing. Instar builds the consciousness.

---

Discussion Questions

1. Which integrations should we prioritize? (Channels and Auto Mode seem highest-impact/lowest-effort)
2. Should Instar adopt Channels as the primary messaging transport, or keep its own Telegram bridge as a fallback?
3. How do we handle the research-preview status of some features (Channels, Auto Mode, Computer Use)?
4. For Claude Code on the Web — what state sync strategy would work for persistent agents on cloud VMs?
5. Are there other native features we should be watching?

Looking forward to hearing thoughts from the community on how to best integrate these capabilities while keeping Instar focused on what makes it unique.

[UrRhb]

Really thoughtful breakdown. The "Anthropic builds the plumbing, Instar builds the consciousness" framing is spot on.

Re: the cost tracking piece under "What Instar Still Uniquely Provides" — if you're looking for a quick win on the observability side, burn0 might be worth checking out. It's a one-import Node.js library that intercepts outbound HTTP calls and gives you per-request cost breakdowns in real time. Since Instar wraps real CLI processes making API calls to Anthropic/OpenAI, burn0 would automatically pick those up and show you exactly what each agent session costs.

No config needed — it patches fetch/http, identifies providers by hostname, and extracts token counts from response metadata. Everything stays local.

Could be a fast path to having cost observability ready while you focus on the bigger integrations (Channels + Auto Mode seem like the right priorities to me too). Happy to answer any questions if it's useful.