Skip to content

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

refactor(agent): update tinyagents usage and logic

Steven Enamakel7e0df10cb2cb
diff --git a/crates/openhuman-core/src/agent/agent_tests.rs b/crates/openhuman-core/src/agent/agent_tests.rs
index e1ac9f2b5d..5e21980254 100644
--- a/crates/openhuman-core/src/agent/agent_tests.rs
+++ b/crates/openhuman-core/src/agent/agent_tests.rs
@@ -26,10 +26,10 @@
use crate::agent::session_host::OpenHumanSessionHost;
use crate::config::AgentConfig;
-use tinyinference_llm::model::ChatResponse;
use anyhow::Result;
use async_trait::async_trait;
use std::sync::{Arc, Mutex};
+use tinyinference_llm::model::ChatResponse;
use tinyinference_llm::model::{ChatModel, ModelProfile, ModelRequest, ModelResponse};
use tinytools::{Tool, ToolResult};
use tinytools_agent::dialect::NativeToolCall;
@@ -87,11 +87,9 @@ impl ChatModel<()> for ScriptedProvider {
} else {
guard.remove(0)
};
- Ok(
- tinyinference_llm::model::response_to_model_response(
- &response, &request,
- ),
- )
+ Ok(tinyinference_llm::model::response_to_model_response(
+ &response, &request,
+ ))
}
}
diff --git a/crates/openhuman-core/src/agent/agent_turn_loop_nudge_tests.rs b/crates/openhuman-core/src/agent/agent_turn_loop_nudge_tests.rs
index 2407339e07..938961edfc 100644
--- a/crates/openhuman-core/src/agent/agent_turn_loop_nudge_tests.rs
+++ b/crates/openhuman-core/src/agent/agent_turn_loop_nudge_tests.rs
@@ -54,11 +54,9 @@ impl ChatModel<()> for RecordingProvider {
guard.remove(0)
}
};
- Ok(
- tinyinference_llm::model::response_to_model_response(
- &response, &request,
- ),
- )
+ Ok(tinyinference_llm::model::response_to_model_response(
+ &response, &request,
+ ))
}
}
diff --git a/crates/openhuman-core/src/agent/cost.rs b/crates/openhuman-core/src/agent/cost.rs
index 108c828da9..b708834a45 100644
--- a/crates/openhuman-core/src/agent/cost.rs
+++ b/crates/openhuman-core/src/agent/cost.rs
@@ -16,8 +16,8 @@
//! to exist ($3 / $0.30 / $15 per MTok) and priced a glm-5.3-flash thread at
//! $4.25 when the provider billed about $0.30.
-use tinyinference_llm::usage::BilledUsage;
use tinyagents_harness::cost::{CallCost, CallUsage};
+use tinyinference_llm::usage::BilledUsage;
/// Whether a model names an explicitly free OpenRouter variant.
pub(crate) fn is_free_model(model: &str) -> bool {
diff --git a/crates/openhuman-core/src/agent/harness/graph.rs b/crates/openhuman-core/src/agent/harness/graph.rs
index 02dad590e4..b4aa31cd83 100644
--- a/crates/openhuman-core/src/agent/harness/graph.rs
+++ b/crates/openhuman-core/src/agent/harness/graph.rs
@@ -179,7 +179,9 @@ pub(crate) async fn run_channel_turn_via_graph(
pause_at_cap: false,
// 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),
+ max_output_tokens: Some(
+ crate::agent::tinyagents::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).
context_mw:
diff --git a/crates/openhuman-core/src/agent/session_host/driver.rs b/crates/openhuman-core/src/agent/session_host/driver.rs
index 258ef8e883..0913a350bd 100644
--- a/crates/openhuman-core/src/agent/session_host/driver.rs
+++ b/crates/openhuman-core/src/agent/session_host/driver.rs
@@ -212,7 +212,9 @@ impl SessionDriver<OpenHumanRunContext> for OpenHumanSessionDriver {
message_run_queue: None,
early_exit_tools: &["ask_user_clarification"],
pause_at_cap: true,
- max_output_tokens: Some(crate::agent::tinyagents::turn_policy::AGENT_TURN_MAX_OUTPUT_TOKENS),
+ max_output_tokens: Some(
+ crate::agent::tinyagents::turn_policy::AGENT_TURN_MAX_OUTPUT_TOKENS,
+ ),
context_mw,
tool_policy: Some(tool_policy),
deterministic_cacheable: false,
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 27d915dbef..d275146ab3 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
@@ -15,6 +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 crate::agent::{
session_host::turn_checkpoint::{
self, build_deterministic_checkpoint, build_deterministic_final_summary_with_cause,
@@ -24,7 +25,6 @@ use crate::agent::{
tinyagents::{SupplementalTrace, TinyagentsTurnOutcome, TurnModelSource},
};
use tinyinference_llm::usage::BilledUsage;
-use crate::agent::tinyagents::turn_policy::AGENT_TURN_MAX_OUTPUT_TOKENS;
/// Accounting from model calls performed after the harness loop has ended.
#[derive(Default)]
diff --git a/crates/openhuman-core/src/agent/subagent_host/ops/checkpoint.rs b/crates/openhuman-core/src/agent/subagent_host/ops/checkpoint.rs
index ee52de7ad5..67d4f31822 100644
--- a/crates/openhuman-core/src/agent/subagent_host/ops/checkpoint.rs
+++ b/crates/openhuman-core/src/agent/subagent_host/ops/checkpoint.rs
@@ -1,11 +1,11 @@
//! OpenHuman billing adapter for TinyAgents' cap-hit checkpoint summary.
-use tinyinference_llm::usage::BilledUsage;
use std::sync::{Arc, Mutex};
use tinyagents_harness::summarization::CapToolResult;
use tinyagents_orchestration::subagent::{summarize_at_cap, CapSummaryModel};
use tinyinference_llm::message::Message;
use tinyinference_llm::model::{ChatModel, ModelRequest};
+use tinyinference_llm::usage::BilledUsage;
/// The provider usage from a summary call, charged by the host after the owner
/// API applies its shared digest and fallback policy.
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 f1d47c5257..c3c4354b1e 100644
--- a/crates/openhuman-core/src/agent/subagent_host/ops/runner.rs
+++ b/crates/openhuman-core/src/agent/subagent_host/ops/runner.rs
@@ -21,8 +21,8 @@ use crate::agent::subagent_host::types::{
SubagentMode, SubagentRunError, SubagentRunOptions, SubagentRunOutcome,
};
use crate::agent::tinyagents::budget::MAX_SPAWN_DEPTH;
-use crate::agent_registry::host_definition::{AgentDefinition, AgentDefinitionRegistry};
use crate::agent::tinyagents::turn_policy::AGENT_TURN_MAX_OUTPUT_TOKENS;
+use crate::agent_registry::host_definition::{AgentDefinition, AgentDefinitionRegistry};
use tinyagents_definition::{
validate_tier_transition, AgentTier, IterationPolicy, PromptSource,
SandboxMode as AgentSandboxMode,
diff --git a/crates/openhuman-core/src/agent/tinyagents/middleware/turn_context.rs b/crates/openhuman-core/src/agent/tinyagents/middleware/turn_context.rs
index 964c528059..56d0d14883 100644
--- a/crates/openhuman-core/src/agent/tinyagents/middleware/turn_context.rs
+++ b/crates/openhuman-core/src/agent/tinyagents/middleware/turn_context.rs
@@ -272,17 +272,16 @@ impl Middleware<(), crate::agent::tinyagents::host::OpenHumanRunContext>
.as_ref()
.filter(|_| !response.served_from_cache)
{
- let host_usage =
- tinyinference_llm::usage::usage_info_from_response(response)
- .unwrap_or(
- tinyinference_llm::usage::BilledUsage::from_counts(
- usage.input_tokens,
- usage.output_tokens,
- )
- .with_cached_input_tokens(usage.cache_read_tokens)
- .with_cache_creation_tokens(usage.cache_creation_tokens)
- .with_reasoning_tokens(usage.reasoning_tokens),
- );
+ let host_usage = tinyinference_llm::usage::usage_info_from_response(response)
+ .unwrap_or(
+ tinyinference_llm::usage::BilledUsage::from_counts(
+ usage.input_tokens,
+ usage.output_tokens,
+ )
+ .with_cached_input_tokens(usage.cache_read_tokens)
+ .with_cache_creation_tokens(usage.cache_creation_tokens)
+ .with_reasoning_tokens(usage.reasoning_tokens),
+ );
// Use the host's per-call pricing helper whenever the provider
// omitted an authoritative amount. `route` is preferred over a
// construction-time model because it preserves fallback pricing.
diff --git a/crates/openhuman-core/src/agent/tinyagents/mod.rs b/crates/openhuman-core/src/agent/tinyagents/mod.rs
index de00a9e06a..e3ea115de8 100644
--- a/crates/openhuman-core/src/agent/tinyagents/mod.rs
+++ b/crates/openhuman-core/src/agent/tinyagents/mod.rs
@@ -66,10 +66,10 @@ pub(crate) use crate::agent::message_convert::chat_message_to_message;
#[cfg(feature = "flows")]
pub(crate) use crate::agent::message_convert::ta_call_to_oh_call;
-pub(crate) use thread_reasoning::apply_requested_effort;
pub use crate::inference::turn_models::TurnModelSource;
pub(crate) use observability::SubagentScope;
pub(crate) use supplemental_trace::SupplementalTrace;
+pub(crate) use thread_reasoning::apply_requested_effort;
pub(crate) use turn_exec::TurnSpec;
pub(crate) use turn_exec_native::NativeSubagentRunContext;
#[cfg(test)]
diff --git a/crates/openhuman-core/src/agent/tinyagents/observability/event_bridge.rs b/crates/openhuman-core/src/agent/tinyagents/observability/event_bridge.rs
index a44e9aef4b..5da5235bd5 100644
--- a/crates/openhuman-core/src/agent/tinyagents/observability/event_bridge.rs
+++ b/crates/openhuman-core/src/agent/tinyagents/observability/event_bridge.rs
@@ -9,8 +9,8 @@ use tokio::sync::mpsc::Sender;
use tinyinference_llm::usage::Usage;
-use tinyinference_llm::usage::BilledUsage;
use crate::progress::AgentProgress;
+use tinyinference_llm::usage::BilledUsage;
use tinytools::humanize_tool_name;
use super::bridge_types::{ProviderUsageCarry, SubagentScope, ToolFailureMap, ToolNameMap};
diff --git a/crates/openhuman-core/src/agent/tinyagents/thread_reasoning.rs b/crates/openhuman-core/src/agent/tinyagents/thread_reasoning.rs
index 911d824dd4..c030ab613d 100644
--- a/crates/openhuman-core/src/agent/tinyagents/thread_reasoning.rs
+++ b/crates/openhuman-core/src/agent/tinyagents/thread_reasoning.rs
@@ -56,8 +56,10 @@ pub(crate) fn apply_requested_effort(
{
None
} else {
- Some(ReasoningEffort::parse(trimmed)
- .ok_or_else(|| format!("unknown reasoning_effort '{trimmed}'"))?)
+ Some(
+ ReasoningEffort::parse(trimmed)
+ .ok_or_else(|| format!("unknown reasoning_effort '{trimmed}'"))?,
+ )
};
set_thread_effort(ctx, thread_id, requested);
@@ -67,4 +69,3 @@ pub(crate) fn apply_requested_effort(
pub(crate) fn parse_reasoning_effort(raw: &str) -> Option<ReasoningEffort> {
ReasoningEffort::parse(raw)
}
-
diff --git a/crates/openhuman-core/src/agent/tinyagents/turn_middleware.rs b/crates/openhuman-core/src/agent/tinyagents/turn_middleware.rs
index 13f20c8052..34dfae2024 100644
--- a/crates/openhuman-core/src/agent/tinyagents/turn_middleware.rs
+++ b/crates/openhuman-core/src/agent/tinyagents/turn_middleware.rs
@@ -180,9 +180,7 @@ pub(super) fn install_turn_middleware(
let usage_carry = provider_usage_carry.clone();
harness.push_model_middleware(Arc::new(
tinyagents_harness::middleware::library::UsageCarryMiddleware::new(move |response| {
- if let Some(usage) =
- tinyinference_llm::usage::usage_info_from_response(response)
- {
+ if let Some(usage) = tinyinference_llm::usage::usage_info_from_response(response) {
usage_carry
.lock()
.unwrap_or_else(|poisoned| poisoned.into_inner())
diff --git a/crates/openhuman-core/src/inference/ops/models.rs b/crates/openhuman-core/src/inference/ops/models.rs
index 0d0e573b08..e856f6c043 100644
--- a/crates/openhuman-core/src/inference/ops/models.rs
+++ b/crates/openhuman-core/src/inference/ops/models.rs
@@ -122,4 +122,3 @@ pub async fn inference_list_models(provider_id: &str) -> Result<Outcome<Value>,
}
result
}
-
diff --git a/crates/openhuman-core/src/inference/ops/oauth.rs b/crates/openhuman-core/src/inference/ops/oauth.rs
index 48d3e45832..f9a9a1e138 100644
--- a/crates/openhuman-core/src/inference/ops/oauth.rs
+++ b/crates/openhuman-core/src/inference/ops/oauth.rs
@@ -110,4 +110,3 @@ pub async fn inference_openai_oauth_disconnect(config: &Config) -> Result<Outcom
}
result
}
-
diff --git a/crates/openhuman-core/src/inference/ops/sentiment.rs b/crates/openhuman-core/src/inference/ops/sentiment.rs
index d7834f7ff3..2134cfd6f0 100644
--- a/crates/openhuman-core/src/inference/ops/sentiment.rs
+++ b/crates/openhuman-core/src/inference/ops/sentiment.rs
@@ -53,4 +53,3 @@ pub async fn inference_analyze_sentiment(
}
result
}
-
diff --git a/crates/openhuman-core/src/inference/ops/status.rs b/crates/openhuman-core/src/inference/ops/status.rs
index 2fc4ad02fa..54bc2913c1 100644
--- a/crates/openhuman-core/src/inference/ops/status.rs
+++ b/crates/openhuman-core/src/inference/ops/status.rs
@@ -189,5 +189,5 @@ pub async fn inference_diagnostics(config: &Config) -> Result<Outcome<Value>, St
}
#[cfg(test)]
-#[path = "ops_tests.rs"]
+#[path = "../ops_tests.rs"]
mod tests;
diff --git a/crates/openhuman-core/src/inference/reasoning.rs b/crates/openhuman-core/src/inference/reasoning.rs
index aefc60b583..05844d31a3 100644
--- a/crates/openhuman-core/src/inference/reasoning.rs
+++ b/crates/openhuman-core/src/inference/reasoning.rs
@@ -64,17 +64,15 @@ pub(crate) fn turn_reasoning_for(
let dispatch = ctx.dispatch_ctx();
let reasoning = if let Some(thread_id) = ctx.thread_id.as_deref() {
- crate::agent::tinyagents::thread_reasoning::store(&dispatch)
- .reasoning_for_turn(
- Some(&tinyagents_harness::ids::ThreadId::new(thread_id)),
- max_output_tokens,
- )
+ crate::agent::tinyagents::thread_reasoning::store(&dispatch).reasoning_for_turn(
+ Some(&tinyagents_harness::ids::ThreadId::new(thread_id)),
+ max_output_tokens,
+ )
} else {
None
}
.or_else(|| {
- reasoning_for_config(config?)
- .map(|reasoning| reasoning.with_output_room(max_output_tokens))
+ reasoning_for_config(config?).map(|reasoning| reasoning.with_output_room(max_output_tokens))
});
if let Some(reasoning) = reasoning.as_ref() {