Dispatch The First Agentic Integrated Overview Environment (IOE)
Stop typing code. Start directing agents. Dispatch orchestrates specialized AI coding agents across a structured pipeline with human review gates.
/Users/jason/code/api-service
index.ts
auth.ts
package.json
CLAUDE.md
tsconfig.json
Auto Model Routing
Implementation tasks routed per-task. Spec and verify use the profile.
Pipeline Profile
Standard (default)
Optimization
Auto (balanced)
Pipeline: Spec → Implement → Review → Verify
Cmd+K Navigate Cmd+M Mission
A View Pipeline Pipeline DAG and progress
N Quick Launch One-off agent task (Cmd+N)
S Sessions Import from Claude, Codex, Gemini
R Import Paper arXiv, URL, or paste abstract
Spec 1
Gemini SECURITY
OWASP scan: auth endpoints
Checking injection vectors on /api/auth/*...
Steer agent
Implement 2
Claude IMPLEMENT
JWT verification middleware
Writing src/middleware/auth.ts...
Steer agent
SWARM 3 agents / file-boundary RUNNING
2/3 agents active
Frontend 1
Qwen Coder CLOUD FRONTEND
Generate API client SDK
Building typed client from OpenAPI spec...
Steer agent
Review 1
HITL REVIEW
Rate limiting implementation
Ready for human review
Approve Reject
GOAL Add JWT auth middleware Running
SPEC Auth spec Done
TEST SPEC Test coverage Done
SEC SPEC OWASP audit Running
IMPLEMENT auth.ts Claude
VERIFY test suite Codex
REVIEW human gate HITL
3D Floor requires WebGL. Enable hardware acceleration in your browser settings.
PIPELINE SPEC → TEST → SECURITY → IMPLEMENT → FRONTEND → VERIFY → REVIEW → AUDIT
AGENT ACTIVITY
Claude / JWT middleware RUNNING
Gemini / OWASP scan RUNNING
Codex / Rate limiter REVIEW
GOALS
JWT auth middleware 43%
DB refactor 71%
QUICK ACTIONS
NEW MISSION K
H Toggle HUD F Fullscreen
M Mission ESC Back
1-8 Focus stage R Review
POOL
2 TEST 5
ACTIVE QUEUED CAPACITY
TOTAL SLOTS 2/5
CLAUDE 1/3
GEMINI 1/2
CODEX 0/2
SECURITY
SECURITY
0 SECURITY 0 WARN 0 FAIL
now-8m now-6m now-4m now-2m now
Claude
Generate auth spec
JWT verification middleware
Active Plans + New Plan
Auth system rollout Running
3 goals · 12 tasks · Started 2h ago
JWT middleware Done
OWASP audit Running
Rate limiting Pending
DB refactor sprint Scheduled
5 goals · 20 tasks · Starts tomorrow
Abstract
The dominant sequence transduction models are based on complex recurrent or convolutional neural networks. We propose a new simple network architecture, the Transformer, based solely on attention mechanisms. Experiments on two machine translation tasks show these models are superior in quality while being more parallelizable and requiring significantly less training time.
Use as context for next goal
Sessions
Import conversations from Claude, Codex, or Gemini to continue as Dispatch goals.
Claude Auth middleware discussion 2h ago
14 messages · 22.4k tokens
Import as goal Codex DB schema design Yesterday
8 messages · 11.1k tokens
Import as goal Gemini Security review notes 2 days ago
6 messages · 8.3k tokens
Import as goal Activity Log
now Claude writing src/middleware/auth.ts
1m ago Gemini flagged XSS risk in token storage
3m ago HITL review approved: rate limiting implementation
4m ago Codex completed test spec generation
6m ago Claude completed auth spec
8m ago Goal dispatched: Add JWT auth middleware
Findings Secrets History Runtime/EDR Trust
Gate Status Pass No issues
Secrets Detected 0 Clean
OWASP Findings 1 XSS flag
Destructive Ops 0 None blocked
Recent Findings SEC XSS risk: token stored in localStorage. Recommend httpOnly cookie. 2m ago
OK No hardcoded secrets detected in auth.ts 3m ago
OK RS256 algorithm properly configured, no weak ciphers 4m ago
Replays Missions
JWT auth pipeline 12 decisions · 4m ago
DB refactor 8 decisions · 1h ago
Rate limiter 5 decisions · 2h ago
JWT auth pipeline - Decision Replay SPEC Generated auth spec with RS256 + refresh token rotation 4m ago
TEST Created 14 test cases covering token expiry, rotation, revocation 3m ago
SEC Flagged: token in localStorage (XSS risk). Recommended httpOnly cookie. 2m ago
IMPL Writing middleware with httpOnly cookie per security spec now
Overview Activity & Permissions Dark Code
Total Tokens 142.8k +12% efficiency
Est. Cost $1.42 3 goals today
Agents Active 3/3 All healthy
Review Queue 1 Awaiting approval
AGENT CONNECTIONS
Local CLI adapters detected on this machine.
Claude connected
claude v1.0.18
/usr/local/bin/claude
Gemini connected
gemini v0.3.6
/usr/local/bin/gemini
Codex connected
codex v0.1.2
/usr/local/bin/codex
CUSTOM AGENTS
Add any CLI tool as an adapter. Define command, args, and capabilities.
No custom agents yet + Add Agent
PIPELINE PROFILES
Configure which agents handle each pipeline stage.
Cloud models, usage, and subscription.
Gemma 4 26B, Qwen Coder 32B, Qwen 3 30B, DeepSeek R1 32B, Llama 3.3 70B + 3 more
Gemma 4 Qwen Coder Qwen 3 Llama 3.3 70B DeepSeek R1 Llama 4 Scout Kimi K2.5 GPT-OSS 20B Mistral Small
Manage Subscription Sign Out
GITHUB
Repository access and pull request integration.
github.com/jason
REMOTE HOSTS
SSH hosts for remote agent execution.
No remote hosts configured + Add Host
DESKTOP
Window behavior and system integration.
3/5 1 reviews 2 team 142k tok 0 err
CL 42k (12) DC 42% AUTO PRO 34% HTTPS main Output Diff Spec
Analyzing security requirements from spec...
Identified 3 core constraints from sec-spec.
Token storage: httpOnly cookie (per security flag)
> Writing src/middleware/auth.ts
import { verify } from 'jsonwebtoken';
import { getCookie } from '../utils/cookie';
export const verifyJWT = async (req, res, next) => {
const token = getCookie(req, 'auth_token');
if (!token) return res.status(401).json({
error: 'Authentication required'
});
try {
const decoded = verify(token, process.env.JWT_SECRET, {
algorithms: ['RS256']
});
req.user = decoded;
next();
} catch (err) {
return res.status(401).json({
error: 'Invalid token'
});
}
};
> Implementing token refresh rotation...
Installed Adapters
claude 2.1.104 (Claude Code)
gemini 0.37.1
codex codex-cli 0.120.0
Cloud Models PRO
Gemma 4 26B
Qwen Coder 32B
Qwen 3 30B
DeepSeek R1 32B
Llama 3.3 70B
Llama 4 Scout
Kimi K2.5
GPT-OSS 20B
Mistral Small
Runtime
Server 0.1.5
Mode local + cloud
Uptime 42m 18s
Platform macOS (arm64)
Command palette search M New Mission Dispatch a quick mission to any agent (Cmd+M)
R Import Research Paper Import from arXiv or paste an abstract
Rp Papers Library Browse imported papers, status tracking, linked goals
F Floor Live agent monitoring
3D Floor (3D) Immersive agent monitoring
A Agentflow Pipeline DAG view
T Task Board Kanban-style task overview
P Plans Goal decomposition and execution plans
V Vault Replay files and snapshots
O Observatory Metrics, costs, insights
X Sessions Import external agent sessions
L Activity Searchable log of all agent actions
R Performance Adapter metrics and correction rules
Sc Security Security findings, secret detection, gate status
Adapter Claude
Mode Standard
Model Auto
QUALITY PIPELINE 6 gates between agents and your codebase
Every change passes 6 quality gates before it reaches your codebase. Tests before code. Security scanning. Clean code enforcement. Design quality. Performance checks. Human review.
Agents follow industry best practices for your domain out of the box -- from auth patterns to database design. No instruction files needed.
Spec
Tests
Security
Clean Code
Performance
Review
Learn how the pipeline works →
ORCHESTRATION Right agent, right job Claude, Codex and Gemini. Auto-route each pipeline stage to the best agent based on task type and model strengths.
SECURITY Secure by default OWASP-mapped specs, destructive command detection and dual-agent security audits before any code ships.
AGENT SWARM Parallel power Deploy multiple agents on a single goal. Each works on a separate part, then changes are merged and verified together.
VISIBILITY See everything, miss nothing Dashboard, Floor swimlanes, pipeline DAG, task board, cost observatory. Every view designed for a different decision.
D DashboardF FloorA AgentflowO Observatory
MEMORY Full decision trail Every agent decision, spec revision and human override is recorded in the Vault. Replay any pipeline from start to finish.
PRIVACY Local-first, always Everything runs on your machine. SQLite database, no cloud dependency, no telemetry by default.
GITHUB PRs and issues, connected Auto-create pull requests from completed goals. Import GitHub issues as goals. Review comments flow back as agent feedback.
VERIFICATION Adversarial cross-review Multiple independent verifiers review every change. You only see the disputes. Consensus findings are auto-resolved.
RESILIENCE Self-healing agents Crash recovery, checkpoint resume and cross-agent state transfer. Agent sessions survive crashes, sleep and network drops.
ROUTING Smart task routing Dispatch learns which agents perform best in your codebase and routes tasks accordingly. Optimize for speed, cost or quality.
CLOUD TOOLS Server-side tool execution Cloud models run file operations in ephemeral workspaces. No round-trips, automatic cleanup, secret scanning before upload.
Pro