The platform, unabridged.
Dap spans the full SDLC. Here's how it works.
Integrations
Source control
Picks up issues, opens PRs, and posts results back to the originating ticket.
Issue tracker
Issues transition automatically; triggered sessions link back to the ticket.
Error monitoring
Error alerts trigger an agent session automatically, no manual triage needed.
Claude Code
The agents that run inside Dap's isolated workspaces.
Codex
The same workspace-native experience as Claude Code.
- [ AND MORE ]
Lifecycle Features
01. Agents activate when work appears — no human trigger required.
Work arrives from GitHub, Jira, error monitoring, or CVE feeds — agents respond without waiting for a human to start them.
- > New feature
Triggered from GitHub issues or Jira tickets automatically.
- > Bug report
Triggered from error monitoring tools (AppSignal, Bugsnag, Sentry).
- > Security issue
Triggered from CVE feeds and Dependabot vulnerability alerts.
31.6%
unattended sessions
02. Every task starts with a fully configured environment. Zero setup.
Every task gets its own isolated workspace — provisioned automatically and ready before the session begins.
- > Workspace orchestration
Isolated environments, pod or inline process, one per task.
- > Warm environments
Ready before the session begins; the same team conventions for every agent and every developer.
- > Workspace port access
Expose workspace ports securely for live in-workspace previews.
10s
median workspace ready time
03. Agents reason, code, and test — you watch or step in.
Choose your AI, follow every decision as it happens, and guide the agents whenever you need to.
- > Choose your AI
Run a configurable squad of agents orchestrated by Dap Harness, or switch to your own subscription (Claude Code, Codex, OpenCode, or similar).
- > Live session streaming
Every reasoning step and tool call streamed as it happens.
- > Attended or unattended
Send instructions, course-correct, or intervene mid-session when you want to, or let the agent see the task through on its own.
4.7×
cheaper than Claude API — per million tokens
04. Every session makes the next one better.
Dap connects your tools and feeds their context to agents — so each session builds on what came before.
- > Context layer
Session history, Jira, GitHub, docs, and CI aggregated into a persistent knowledge graph; relevant context pushed into every session before it begins.
- > MCP integrations
Connect databases, internal APIs, and custom tools via authorized MCP servers; agents access them without unrestricted access.
- > Errors sharpen workflows
Failed sessions categorized and fed back into the knowledge graph, making every future run smarter.
05. AI development becomes a team activity.
No more black-box sessions — hand off work, share progress, and preview results with your team.
- > Hand off to a colleague
Transfer an active session to another engineer in real time.
- > Share sessions
Grant read or write access to any contributor at any point.
- > Share preview environments
Share live workspace URLs for immediate feedback.
06. No offenders before the PR opens.
Automated review runs until the code meets the bar — human gate or auto-merge, your call.
- > Automated code reviews
Dap picks up code review feedback (human reviews or your own AI code review providers) and drives fixes automatically.
- > Autofix review feedback
Agent addresses review comments and reruns until no offenders remain.
- > Human review gate
You decide where the gate sits: human approval, or auto-merge once the AI jury's confidence score clears the bar.
3.3×
faster delivery — auto-merge vs. human-gated
07. Connected to the pipeline your team already uses.
Dap plugs into your CI system and surfaces results to whoever needs them — engineers, testers, or product.
- > Hooks into your existing CI
Every agent PR triggers your CI pipeline automatically, no new setup.
- > Autofix CI failures
Dap subscribes to your CI's events; on failure, the agent diagnoses the issue, fixes it, and reruns.
- > Provision preview environments
QA, product, or stakeholders can generate one for review.
08. One AI jury. One confidence score. Ships when ready.
An AI jury assesses every PR before it is eligible to merge — low-risk changes merge automatically, everything else waits for human approval.
- > Merge policy you control
AI jury criteria and confidence thresholds configurable per repository.
- > Reasoning built into the PR
Every action, decision, and file change documented back to the PR and originating ticket.
- > Changelogs & notificationsComing soon
Teams are notified and changelogs generated automatically on delivery.
37%
of eligible PRs auto-merge
Multi-agent harness & A2A
The A2A protocol runs the whole thing. Dap built its harness in-house — for performance and for cost control, running each agent on the model best suited to its role. The harness coordinates the squad; the protocol opens the door to everything else.
Multi-agent harness — internal squad orchestration
- Squad orchestration — the harness coordinates execution
- Agents talk over A2A internally — even squad-internal coordination speaks the open protocol, not a proprietary message bus
A2A interoperability — beyond Dap
- Built on open protocol — A2A is an open standard; as the ecosystem grows, Dap's interoperability grows with it, no rebuild required
- Bidirectional — your Dap agents reach any external A2A-compliant agent, and external agents reach yours. Both directions, same protocol.
cheaper than Claude API — per million tokens
Apps
Dap is not just an orchestration layer — it's also where you work. Apps bring the tools developers already use directly into the Dap interface, so the agentic session and the development environment live side by side. This list isn't fixed — new Apps get added as the platform grows.
Dap Agentic Livestream
The primary window into agent execution: every reasoning step, tool call, and decision streamed live as it happens.
Claude Code
Dap's native agent experience. Bring your own subscription — no separate API keys, no workarounds, no breaking terms of service.
Codex
For teams standardized on OpenAI. Bring your own subscription and get the same workspace-native experience.
Grok
[COMING SOON]If you're a bit more edge, bring your own subscription — no separate API keys, no workarounds, no breaking terms of service.
Terminal
Direct terminal access to the workspace. Run commands, inspect the environment, or intervene mid-session without leaving Dap.
Hunk
Interactive diff and patch tool for reviewing and staging changes directly inside the workspace.
NeoVim
Full NeoVim editor embedded in the Dap interface for developers who prefer keyboard-driven editing.
Visual Studio
Visual Studio integration for teams that work in a GUI-first editor environment.
Cross-cutting features
09. Contained by design. Safe by architecture.
Security isn't a setting — it's built into how every workspace is provisioned and destroyed. Damage is limited by design.
- > Bounded blast radius
Dedicated, isolated workspaces with no shared compute; agents work on branches, never main.
- > Never leak a secret
Credentials are scoped to environments and never visible to agents; they can't leak what they don't have access to.
- > Deployment
Run in your own VPC, or fully on-premise for teams that can't put code on someone else's infrastructure.
10. You set the rules. The platform enforces them.
Know what agents can access, what they did, and what they're allowed to do — before and after every session.
- > Common toolset
Agents operate within the same boundaries as your developers; no hidden capabilities.
- > Full audit trail
Every action, reasoning step, and file change is logged.
- > Policy enforcement
Per-repository rules govern what agents can access, run, and touch.
11. Full visibility into what the platform is doing.
Logs, metrics, and dashboards — built in, not bolted on.
- > Logs
Full session and workspace logs, structured and searchable across every task.
- > MetricsComing soon
Data across sessions, workspaces, and agents.
- > Custom DashboardsComing soon
Pre-built dashboards for session lifecycle, workspace provisioning, and event flow.
12. Measure what AI is actually delivering.
Coming soonTrack session outcomes, delivery speed, and where humans need to step in — over time.
- > Session performance
Success rates, failure patterns, time from session start to PR open.
- > Delivery throughput
PRs shipped per agent, per repository, per team.
- > Human vs AI intervention
Where agents complete work autonomously vs. where humans step in.
13. Reviews that run without you asking.
Coming soonDap reviews architecture, code quality, and processes on a schedule — independently of any session.
- > Architecture Reviews
Periodic AI review of decisions and patterns across the codebase.
- > Quality Reviews
Coverage gaps, complexity hotspots, and technical debt, surfaced automatically.
- > Process Reviews
Surface inefficiencies and improvement opportunities across your workflow.