Implementation comparison
Compare the runtime.
Inspect the evidence.
Compare how agent projects handle durable execution, permissions, costs, and operational visibility.
Code snapshot inspected August 3, 2026. Capabilities may have changed since this audit.
01
The code-only comparison
What does the implementation provide?
This retained audit describes the inspected code and versions below. “Not found” means the audit did not establish the capability. Closed components are marked separately; external dependencies are shown explicitly.
- Implemented in inspected code
- Partial or different scope
- External dependency
- Not established in inspected code
- Not inspectable in this audit
| Production capability | Cayu | LangGraph | OpenAI Agents | PydanticAI | Claude Agent SDK | CrewAI |
|---|---|---|---|---|---|---|
| Crash resume | Implemented in inspected code: ✓ Built in | Partial or different scope: Graph-state resume | Not established in inspected code: Manual RunState only | External dependency: External engine | Partial or different scope: Transcript | Partial or different scope: Flow checkpoints |
| Ambiguous-effect recovery | Implemented in inspected code: ✓ Built in | Not established in inspected code: Failed node reruns | Not established in inspected code: No pre-effect ledger | Not established in inspected code: Not found | Not inspectable in this audit: Closed CLI | Partial or different scope: Async event |
| Restart-safe approvals | Implemented in inspected code: ✓ Built in | Partial or different scope: Durable interrupt only | Partial or different scope: Caller persists state | External dependency: Deferred / external | Not inspectable in this audit: Closed CLI | Partial or different scope: Checkpointed HITL |
| Runtime tool authority | Implemented in inspected code: ✓ Built in | Not established in inspected code: Application wrapper only | Partial or different scope: Per-tool approval hooks | Partial or different scope: Application | Not inspectable in this audit: Closed CLI | Partial or different scope: Application |
| Dollar budget enforcement | Implemented in inspected code: ✓ Reserve + settle | Not established in inspected code: Not found | Not established in inspected code: Not found | Partial or different scope: Cost limit | Not inspectable in this audit: Closed CLI | Not established in inspected code: Not found |
| Virtual credentials + egress | Implemented in inspected code: ✓ Built in | Not established in inspected code: Not found | Not established in inspected code: No virtual creds / egress | Not established in inspected code: Not found | Not inspectable in this audit: Closed CLI | Not established in inspected code: Not found |
| Isolated workspaces | Implemented in inspected code: ✓ Built in | Not established in inspected code: Not found | Implemented in inspected code: ✓ Docker + remote sandboxes | Not established in inspected code: Not found | Not inspectable in this audit: Closed CLI | Not established in inspected code: Not found |
| Durable artifacts | Implemented in inspected code: ✓ Built in | Not established in inspected code: Not found | Not established in inspected code: Workspace snapshots only | Not established in inspected code: Not found | Not inspectable in this audit: Not in OSS | Partial or different scope: Output files only |
| Environment + runner contracts | Implemented in inspected code: ✓ Built in | Not established in inspected code: Not found | Partial or different scope: Sandbox backends only | External dependency: External engine | Not established in inspected code: Fixed CLI transport | Not established in inspected code: Not found |
| Tasks + dispatch + workers | Implemented in inspected code: ✓ Built in | Not established in inspected code: Not found | Not established in inspected code: Not found | External dependency: External engine | Not inspectable in this audit: Closed CLI | Partial or different scope: Crews + Flows only |
| Context recovery | Implemented in inspected code: ✓ Built in | Not established in inspected code: Application hook only | Partial or different scope: OpenAI compaction only | Partial or different scope: History hooks + compaction | Not inspectable in this audit: Closed CLI | Partial or different scope: Summarization |
| Durable event history | Implemented in inspected code: ✓ Built in | Partial or different scope: State snapshots only | Not established in inspected code: Messages + best-effort logs | External dependency: Messages + external history | Partial or different scope: Transcript mirror | Partial or different scope: Checkpoints + event replay |
| Knowledge + memory stores | Implemented in inspected code: ✓ Built in | Partial or different scope: Memory store only | Partial or different scope: Sandbox-local memory | Partial or different scope: Provider-native only | Partial or different scope: Transcript store; memory in CLI | Partial or different scope: Memory only |
| Forks + subagent lineage | Implemented in inspected code: ✓ Built in | Partial or different scope: Graph forks + subgraphs | Not established in inspected code: Handoffs only | Partial or different scope: Delegation only | Partial or different scope: Session forks + transcripts | Partial or different scope: Checkpoint branches + crews |
| Trajectory evals + replay | Implemented in inspected code: ✓ Built in | Not established in inspected code: Checkpoint re-execution only | Not established in inspected code: Tracing only | Not established in inspected code: Evals; no trajectory replay | Not established in inspected code: Not found | Partial or different scope: Trace evals + task replay |
| Self-hosted control plane | Implemented in inspected code: ✓ Built in | Not established in inspected code: Not found | Not established in inspected code: Not found | Not established in inspected code: Dev web UI | Not established in inspected code: Not found | Not established in inspected code: Flow visualization only |
Generated by Claude Code using Fable 5 with fresh session with no memory or preference files.
Read the audit prompt
Read https://cayu.dev/ and compare Cayu against its competition on the features needed to build long-running agents. Ground the research in code. Do not base anything on marketing pages, only on the code.
Choose for the guarantees your workload needs.
Follow the code, configuration, and deployment boundary behind each claim.Code audit revisions
Cayu f4b1490 · LangGraph b2926a0 · OpenAI Agents 08fa43c · PydanticAI a1defdf · Claude Agent SDK 60ecf63 · CrewAI b10c4ff
02
The failure boundary
Recovery at a difficult boundary.
The tool acted. The result never arrived. What happens next?
- 01
Tool starts
Action authorized. - 02
World changes
External effect. - 03
Worker dies
Result missing. - 04
Cayu recovers
Reconcile. Resume.
A checkpoint tells you where the program was.
Cayu records the evidence needed to recover.Kill the worker.
No silent re-execution.
No invented result.
Deliberate recovery boundary.
Proven in a real process-kill test 03
Choose for the work
Choose a foundation
for your domain.
What to evaluate
Look beyond the first run.
Assess each framework’s defaults, extension points, and execution guarantees against the work your agent needs to do.Building with Cayu
Shape your agent in Python.
Assemble your domain-specific harness using Cayu’s components and defaults, with durable execution integrated into the framework.Open-source Python framework for domain-specific long-horizon agents
Build for the long run.
Own the loop. Continue from evidence. Explain every outcome.
Compared against executable open-source code and tests inspected August 3, 2026. No vendor marketing claims used.