Extensible harness

Deep Agents Code

Extensible LangGraph coding agent with headless runs and optional remote sandboxes.

Status: Active
License: MIT
Evidence: Documented + code-verifiable, 6 sources
Product record checked: 2026-07-28

At a glance

LangChain's terminal coding product built on Deep Agents and LangGraph, with an interactive TUI, non-interactive pipeline runner, custom subagents, MCP, local session checkpoints, persistent memory, approval gates, and integrations for managed remote sandboxes.

Good choice if

Python and LangGraph teams that want a turnkey CLI they can also extend
CI and scripted coding workflows that need turn limits, sessions, and structured non-interactive output
Untrusted repositories when paired with one of the documented remote sandbox backends

Check before choosing

Local execution is host-first and trusts the working directory; project artifacts can influence the agent before the first approval prompt
Isolation is an opt-in remote-backend surface whose provider internals and guarantees sit outside the Deep Agents Code repository
The CLI starts a local LangGraph development server and stores sessions in SQLite, adding operational moving parts compared with a small single-binary agent

See 2 more considerations

MCP servers, hooks, skills, and repository instructions expand the trusted computing base and need their own review
The repository includes project-owned eval packages, but they are not treated as independent comparative evidence or imported into fit scoring

Quick facts

Interfaces: Terminal, Automation
Approval style: Interactive approvals
Model portability: Provider choice
Isolation: Managed sandbox
Agent organization: Delegated subagents
Local model path: Not documentedAbsence of current documentation is not proof that the capability is impossible.

View full technical record

Catalog layer: Coding harness
Harness membership: Qualifies: 4 required criteria evidenced
Product role: Extensible harness
Agent organization: Delegated subagents
Runtime posture: Host-first
Isolation options: Managed sandbox
State model: Persistent memory
Context lifecycle: Persistent memory
Permission posture: Interactive approvals
Verification: Tool-assisted checks
Observability: Structured traces
Recovery: Session resume
Interfaces: Terminal, Automation
Provider posture: Multi-provider
Model portability: Provider choice
Subscription access: Not documented
Enterprise access: Not documented
Local model path: Not documentedAbsence of current documentation is not proof that the capability is impossible.
Claim boundary: Harness capability only

Why it qualifies as a coding harness

This confirms category fit, not product quality. Every required criterion links back to first-party evidence.

Qualifies4 of 4 required criteria evidenced

Adaptive agent loop
Documented
The system repeatedly observes results and chooses the next action instead of following a fixed one-pass graph.
EvidenceDeep Agents Code threat model
Repository tool execution
Documented
The system can use tools to inspect and change a repository or its execution environment.
EvidenceDeep Agents Code threat model
Task-aware context management
Documented
The runtime assembles, updates, compacts, retrieves, or persists task-relevant context while work proceeds.
EvidenceDeep Agents Code overview
Model-independent runtime control
Documented
Permissions, budgets, interruption, policy, or stop controls operate outside the model's own text generation.
EvidenceDeep Agents Code threat model

How it works under the hood

Seven mechanisms mapped from first-party records. These labels describe what the harness provides, not how intelligent its model is.

7/7layers documented

Execution & isolationSandbox availableDocumented mechanism, not a performance score.
Tooling & integrationsExtensible toolsDocumented mechanism, not a performance score.
Context & statePersistent stateDocumented mechanism, not a performance score.
Lifecycle & recoverySession resumeDocumented mechanism, not a performance score.
ObservabilityStructured tracesDocumented mechanism, not a performance score.
VerificationTool-assistedDocumented mechanism, not a performance score.
Governance & permissionsApproval promptsDocumented mechanism, not a performance score.

Public code audit

4/5public artifacts present

Security policy: Not found
CI workflow: Present at inspected commit
Automated tests: Present at inspected commit
Evaluation assets: Present at inspected commit
Contributor documentation: Present at inspected commit

The inspected monorepo contains 1,473 files, 47 workflows, roughly 400 test-like paths, contributor guidance, and extensive project-owned evaluation packages, but no repository security policy. Evals cover the broader Deep Agents stack and do not supply an independent immutable product comparison, so no score is imported.

Inspect commit 43eb196cf7faa993f2fa372dcc1fa65572d8a301, checked 2026-07-28

Measured configurations

No benchmark run passes the full metadata admission policy for this harness yet. Missing data is not scored as zero.

Benchmark policy and all runs

Capability support

Documented first-class product support, checked against the sources below.

External tools (MCP): DocumentedProduct-supported MCP integrationSource · checked 2026-07-28The source establishes the mechanism, not its quality or availability in every mode.
Local models: Not documentedNo first-class support established by the current recordAbsence of current documentation is not proof that the capability is impossible.
Agent parallelism: DocumentedDelegated or parallel agent workflowSource · checked 2026-07-28The source establishes the mechanism, not its quality or availability in every mode.
Runs without an open UI: DocumentedNon-interactive or automation surfaceSource · checked 2026-07-28The source establishes the mechanism, not its quality or availability in every mode.
Browser control: Not documentedNo first-class support established by the current recordAbsence of current documentation is not proof that the capability is impossible.
Isolated execution: OptionalOptional managed remote sandbox backends; local execution trusts the host working directorySource · checked 2026-07-28The source establishes the mechanism, not its quality or availability in every mode.
Undo file changes: Not documentedNo first-class support established by the current recordAbsence of current documentation is not proof that the capability is impossible.

Primary evidence

Each capability claim is tied to a first-party record and a verification date.

Product record checked 2026-07-28

Deep Agents Code overviewInteractive coding workflow, web search, resume, remote sandboxes, persistent memory, skills, headless mode, approvals, and installationOfficial repositorySource checked 2026-07-28 Deep Agents Code threat modelAgent loop, side-effecting tools, pre-approval project reads, MCP, subagents, session persistence, headless execution, hooks, and sandbox boundariesOfficial repositorySource checked 2026-07-28 Deep Agents Code non-interactive runnerProgrammatic prompt execution, streaming events, cancellation, session handling, and non-interactive completion behaviorOfficial repositorySource checked 2026-07-28 Deep Agents Code MCP implementationMCP configuration loading, local and remote server transports, tool filtering, and connection managementOfficial repositorySource checked 2026-07-28 Deep Agents Code sandbox factorySelectable integrations for remote execution environments and the boundary between local and managed sandbox modesOfficial repositorySource checked 2026-07-28 Deep Agents inspected source treePinned monorepo source, tests, CI, contributor documentation, packages, examples, and project-owned evaluation assetsOfficial repositorySource checked 2026-07-28

Ecosystem context

OpenRouter coding CLI agent directory Used to discover Deep Agents Code; all published claims resolve to LangChain-controlled sources. Observed 2026-07-28.