Skip to content

The fastest, cheapest, most efficient open-source agent harness. Run more than 500 agents on a $10 VPS.

refactor(tinyagents): source turn output token cap from harness crate

Steven Enamakel0ea038c3c2fd
diff --git a/crates/openhuman-core/src/agent/harness/graph.rs b/crates/openhuman-core/src/agent/harness/graph.rs
index b4aa31cd83..524e3f0c47 100644
--- a/crates/openhuman-core/src/agent/harness/graph.rs
+++ b/crates/openhuman-core/src/agent/harness/graph.rs
@@ -180,7 +180,7 @@ pub(crate) async fn run_channel_turn_via_graph(
// Bound the model's per-call output (legacy parity — channel turns ran at
// the standard per-turn budget).
max_output_tokens: Some(
- crate::agent::tinyagents::turn_policy::AGENT_TURN_MAX_OUTPUT_TOKENS,
+ tinyagents_harness::turn_policy::AGENT_TURN_MAX_OUTPUT_TOKENS,
),
// Context middlewares: cache-align + default tool-result byte cap (the
// channel path has no session context config to source from).
diff --git a/crates/openhuman-core/src/agent/session_host/driver.rs b/crates/openhuman-core/src/agent/session_host/driver.rs
index 0913a350bd..8cde107fa7 100644
--- a/crates/openhuman-core/src/agent/session_host/driver.rs
+++ b/crates/openhuman-core/src/agent/session_host/driver.rs
@@ -213,7 +213,7 @@ impl SessionDriver<OpenHumanRunContext> for OpenHumanSessionDriver {
early_exit_tools: &["ask_user_clarification"],
pause_at_cap: true,
max_output_tokens: Some(
- crate::agent::tinyagents::turn_policy::AGENT_TURN_MAX_OUTPUT_TOKENS,
+ tinyagents_harness::turn_policy::AGENT_TURN_MAX_OUTPUT_TOKENS,
),
context_mw,
tool_policy: Some(tool_policy),
diff --git a/crates/openhuman-core/src/agent/session_host/driver/grounded_close.rs b/crates/openhuman-core/src/agent/session_host/driver/grounded_close.rs
index d275146ab3..eafa5bef23 100644
--- a/crates/openhuman-core/src/agent/session_host/driver/grounded_close.rs
+++ b/crates/openhuman-core/src/agent/session_host/driver/grounded_close.rs
@@ -15,7 +15,7 @@ use tinytools_agent::dialect::ToolDialect;
use tinyagents_harness::middleware::wrap_harness_instruction;
-use crate::agent::tinyagents::turn_policy::AGENT_TURN_MAX_OUTPUT_TOKENS;
+use tinyagents_harness::turn_policy::AGENT_TURN_MAX_OUTPUT_TOKENS;
use crate::agent::{
session_host::turn_checkpoint::{
self, build_deterministic_checkpoint, build_deterministic_final_summary_with_cause,
diff --git a/crates/openhuman-core/src/agent/subagent_host/ops/runner.rs b/crates/openhuman-core/src/agent/subagent_host/ops/runner.rs
index c3c4354b1e..45d90ece81 100644
--- a/crates/openhuman-core/src/agent/subagent_host/ops/runner.rs
+++ b/crates/openhuman-core/src/agent/subagent_host/ops/runner.rs
@@ -21,7 +21,7 @@ use crate::agent::subagent_host::types::{
SubagentMode, SubagentRunError, SubagentRunOptions, SubagentRunOutcome,
};
use crate::agent::tinyagents::budget::MAX_SPAWN_DEPTH;
-use crate::agent::tinyagents::turn_policy::AGENT_TURN_MAX_OUTPUT_TOKENS;
+use tinyagents_harness::turn_policy::AGENT_TURN_MAX_OUTPUT_TOKENS;
use crate::agent_registry::host_definition::{AgentDefinition, AgentDefinitionRegistry};
use tinyagents_definition::{
validate_tier_transition, AgentTier, IterationPolicy, PromptSource,
diff --git a/crates/openhuman-core/src/agent/tinyagents/payload_summarizer.rs b/crates/openhuman-core/src/agent/tinyagents/payload_summarizer.rs
index f81b02cc08..abb80babb2 100644
--- a/crates/openhuman-core/src/agent/tinyagents/payload_summarizer.rs
+++ b/crates/openhuman-core/src/agent/tinyagents/payload_summarizer.rs
@@ -101,7 +101,7 @@ impl HarnessPayloadSummarizer<OpenHumanRunContext> for SubagentPayloadSummarizer
let thread_id = parent_ctx.data.thread_id.clone();
let max_output_tokens = definition
.max_turn_output_tokens
- .unwrap_or(crate::agent::tinyagents::turn_policy::AGENT_TURN_MAX_OUTPUT_TOKENS);
+ .unwrap_or(tinyagents_harness::turn_policy::AGENT_TURN_MAX_OUTPUT_TOKENS);
let child_context = unary_child_context(
parent_ctx,
&definition.id,
diff --git a/crates/openhuman-core/src/agent/tinyagents/turn_policy.rs b/crates/openhuman-core/src/agent/tinyagents/turn_policy.rs
index 3179dbc664..1ae61c1169 100644
--- a/crates/openhuman-core/src/agent/tinyagents/turn_policy.rs
+++ b/crates/openhuman-core/src/agent/tinyagents/turn_policy.rs
@@ -8,8 +8,6 @@
use tinyagents_harness::runtime::RunPolicy;
-/// Maximum output tokens for one hosted agent model call.
-pub(crate) const AGENT_TURN_MAX_OUTPUT_TOKENS: u32 = 16_384;
use tinyagents_harness::turn_policy::{
local_turn_backstop_secs, run_policy, turn_wall_clock_ms, TurnPolicyOptions,
TurnPolicyOverrides,
diff --git a/crates/openhuman-core/src/inference/reasoning_tests.rs b/crates/openhuman-core/src/inference/reasoning_tests.rs
index 804f08d8ff..cd39f2c669 100644
--- a/crates/openhuman-core/src/inference/reasoning_tests.rs
+++ b/crates/openhuman-core/src/inference/reasoning_tests.rs
@@ -245,7 +245,7 @@ fn root_turn_reasoning_reserves_output_room_with_a_budget() {
let reasoning = turn_reasoning_for(
&ctx,
None,
- Some(crate::agent::tinyagents::turn_policy::AGENT_TURN_MAX_OUTPUT_TOKENS),
+ Some(tinyagents_harness::turn_policy::AGENT_TURN_MAX_OUTPUT_TOKENS),
);
set_thread_effort(&crate::core::runtime::DispatchCtx::ambient(), thread, None);
assert_eq!(