HarnessMatch
Menu

Extensible harness

Letta Harness

Memory-first harness for persistent agents across machines and sessions.

Status
Active
License
Apache-2.0
Evidence
Documented + code-verifiable, 23 sources
Product record checked
2026-07-27
Compare harnessesInspect sources

At a glance

The stateful open-source harness formerly called Letta Code, with git-backed self-editing memory, skills, MCP connectivity, trusted runtime mods, foreground or background subagents, headless execution, schedules, multi-provider access, and selectable local, remote, or managed cloud environments.

Good choice if

  • Long-lived agents that should remember preferences, prior work, and learned skills
  • Multi-agent workflows routed across laptops, remote machines, CI, and managed cloud sandboxes
  • Developers combining subscriptions, BYOK providers, local runtimes, MCP, schedules, and messaging surfaces

Check before choosing

  • Its persistent-memory model is more complex than a disposable repository session
  • The interactive CLI currently starts in unrestricted mode; approval prompts require standard mode or explicit persistent policy rules
  • Local client tools execute with the selected computer's shell, files, credentials, and installed software even when agent state lives in Letta Cloud
See 3 more considerations
  • Managed sandboxing is an optional execution target, not the default boundary for local CLI work
  • The platform supports external MCP tools, but current first-party guidance recommends skills instead of MCP for the Letta app and CLI; external servers remain separate execution and trust boundaries
  • Mods execute as fully trusted code inside the harness process and can add tools, commands, hooks, permission policies, UI, and model providers, so they are not a sandboxed extension mechanism

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.

    EvidenceLetta Code quickstart

  • Repository tool execution

    Documented

    The system can use tools to inspect and change a repository or its execution environment.

    EvidenceClient tool execution model

  • Task-aware context management

    Documented

    The runtime assembles, updates, compacts, retrieves, or persists task-relevant context while work proceeds.

    EvidenceMemFS

  • Model-independent runtime control

    Documented

    Permissions, budgets, interruption, policy, or stop controls operate outside the model's own text generation.

    EvidencePermissions

Membership establishes category fit only. It does not score quality, safety, autonomy, model capability, or benchmark performance. · Read the membership rule.

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 & recoveryManaged recoveryDocumented mechanism, not a performance score.
  • ObservabilityLogs/transcriptsDocumented mechanism, not a performance score.
  • VerificationTool-assistedDocumented mechanism, not a performance score.
  • Governance & permissionsPolicy controlsDocumented mechanism, not a performance score.

Public code audit

3/5public artifacts present
Security policy
Not found
CI workflow
Present at inspected commit
Automated tests
Present at inspected commit
Evaluation assets
Not found
Contributor documentation
Present at inspected commit

The v0.29.4 client and harness tree contains 580 test-like files and eleven workflows with contributor documentation but no root security policy or dedicated product-evaluation suite. Managed Letta API, Cloud, and sandbox implementations are outside this client-source audit.

Inspect commit 286a01d10602eab4a356f2b062e817310f992966, checked 2026-07-27

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 2026-07-27The source establishes the mechanism, not its quality or availability in every mode.
Local models
DocumentedLocal or self-hosted model pathSource · checked 2026-07-27The source establishes the mechanism, not its quality or availability in every mode.
Agent parallelism
DocumentedDelegated or parallel agent workflowSource · checked 2026-07-27The source establishes the mechanism, not its quality or availability in every mode.
Runs without an open UI
DocumentedNon-interactive or automation surfaceSource · checked 2026-07-27The 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
Depends on surfaceManaged cloud sandbox; local tools remain host-executedSource · checked 2026-07-27The 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 2026-07-27
Letta Code 0.29.4 source snapshotImmutable Apache-2.0 client and harness source snapshot, including 580 test-like files and eleven workflows; hosted Letta services remain outside the auditOfficial repositoryLetta Code 0.29.4 releaseStable release identity and dated version boundary used for the auditOfficial announcementLetta Code quickstartCLI, desktop and web surfaces, headless mode, coding-plan access, persistent memory, parallel conversations, and schedulesOfficial docsMemFSGit-backed long-term memory, versioned memory edits, system context, skills, synchronization, and shared memory repositoriesOfficial docsPermissionsUnrestricted default, standard approvals, edit-only mode, persistent allow and deny rules, tool restriction, and memory guardsOfficial docsExecution environmentsLocal host execution, managed cloud sandboxes, remote machines, environment-specific files and credentials, and memory separationOfficial docsCloud sandboxesManaged isolated computers, shell and filesystem scope, lifecycle, local-file separation, and agent-state boundariesOfficial docsSubagentsForeground and background delegation, isolated contexts, persistent-agent delegation, custom tools, skills, and memory scopeOfficial docs
View 15 additional sources
Product and interfaces3 sources
Supported model-provider typesHosted, BYOK, subscription OAuth, Ollama, LM Studio, vLLM, SGLang, OpenRouter, and other provider types exposed by the Letta platformOfficial docsCurrent Letta documentation indexCurrent product map and naming boundary for the Letta Harness, agent SDK, API, execution environments, memory, automation, and customizationOfficial docsCLI referenceInteractive and headless commands, backend selection, model and memory controls, permission modes, session resume, environment routing, and automation flagsOfficial docs
Execution and control1 sources
Client tool execution modelLocal client execution, approval handoff, server-versus-client boundaries, local credentials and resources, and Letta Harness as reference implementationOfficial docs
Automation and extensions6 sources
Agent skillsAgent, project, computer, and bundled skill scopes plus persistence, installation, trust warnings, and direct invocationOfficial docsHeadless modeNon-interactive execution, structured streams, permission control, environment routing, persistent conversations, and tool eventsOfficial docsSchedulesDurable cloud and local schedules, execution targets, cloud-sandbox fallback, limits, timezones, and run historyOfficial docsMCP tool execution modelExternal MCP server registration, transports, authentication, remote execution boundary, attachment to agents, and guidance to prefer skills in the app and CLIOfficial docsGitHub ActionRepository installation, issue and pull-request triggers, @letta-code mentions, workflow setup, authentication, and automated coding or review tasksOfficial docsTrusted runtime modsFully trusted in-process extensions for tools, commands, hooks, permissions, providers, UI, reload behavior, sharing, and explicit trust boundaryOfficial docs
Enterprise and operations1 sources
Harness configurationHierarchical global and project settings, backend and authentication modes, provider configuration, environment selection, permissions, and defaultsOfficial docs
Releases and public code audit4 sources
Memory research lineageFirst-party MemGPT paper that motivates tiered persistent memory; research lineage rather than comparative product-performance evidenceOfficial announcementSleep-time compute researchFirst-party research behind offline memory refinement, used as mechanism evidence rather than a harness benchmarkOfficial announcementPinned CLI MCP implementationCode-verifiable HTTP, SSE, and stdio MCP server commands, headers, authentication tokens, argument parsing, and CLI result handling at v0.29.4Official repositoryHarness changelogVersioned Letta Harness changes, permission defaults, memory, channels, mods, providers, schedules, and documentation-lag warningOfficial announcement